TheNote
Google Cloud Blog Note
GooglePlay AppStore

Google Cloud Blog

cloud.google.com/blog is the official blog of Google Cloud. It provides news, updates, and insights on Google Cloud's products and services, as well as trends and innovations in the cloud computing industry. The blog features articles written by Google Cloud experts, engineers, and thought leaders, covering a wide range of topics such as artificial intelligence, machine learning, data analytics, security, and more. The articles often include technical tutorials, case studies, and best practices, making the blog a valuable resource for developers, IT professionals, and business leaders who use or are interested in Google Cloud. The blog is well-organized, with articles categorized by topic, product, and industry. Visitors can browse the latest articles, search for specific topics, or subscribe to the blog's RSS feed to stay up-to-date with the latest news and updates. Some of the key features of the blog include: - In-depth articles on Google Cloud products and services, such as Google Cloud Platform, Google Cloud Storage, and Google Cloud AI Platform - Technical tutorials and guides on how to use Google Cloud services - Case studies and success stories from Google Cloud customers - Insights and analysis on industry trends and innovations - News and updates on Google Cloud's partnerships and collaborations - Interviews with Google Cloud experts and thought leaders Overall, the Google Cloud blog is a valuable resource for anyone interested in cloud computing, artificial intelligence, and related technologies.
Cloud Blog cloud.google.com
RSS cloudblog.withgoogle.com
RSS Hunter RSS Hunter • Aug 19, 2024

Thread Of Notes

Safely run AI-generated code in Cloud Run sandboxes

Google Cloud has introduced Cloud Run sandboxes, a new feature available in public preview. These sandboxes provide a secure and rapid runtime environment for executing untrusted code, such as AI-generated scripts. Previously, developers relied on complex infrastructure or third-party solutions to achieve similar isolation. Cloud Run sandboxes are designed to be lightweight and can be spawned quickly within existing Cloud Run service instances. Key use cases include LLM code interpreters for data analysis, headless browsers for web scraping, and secure execution of user-submitted code. Enabling sandboxes is as simple as adding a flag during service deployment. Once enabled, a sandbox CLI binary is mounted, allowing applications to spawn sandboxes programmatically. The security architecture emphasizes a zero-trust approach, isolating credentials and environments. Network egress is denied by default, and the filesystem overlay is read-only with temporary write capabilities. Cloud Run sandboxes integrate with the Agent Development Kit and ComputeSDK for simplified code execution. This feature runs on existing allocated resources, incurring no additional cost for users.
https://cloud.google.com/blog/topics/developers-practitioners/google-cloud-run-sandboxes-are-in-public-preview/ cloud.google.com
CdXz5zHNQW_CD5sY2MVyG.gif
RSS Hunter RSS Hunter • Yesterday

Solve harder problems with AlphaEvolve, now available to everyone on Google Cloud

Optimization problems, though challenging, are becoming more manageable with AI. Traditional coding struggles with vast search spaces for optimal algorithms. AlphaEvolve, a code optimization and discovery agent powered by Gemini, addresses this by systematically exploring solutions. It is now generally available on the Gemini Enterprise Agent Platform.The deployment process involves four steps: Define the problem with a baseline algorithm and context; Measure by establishing a scoring function for correctness, performance, and constraints; Optimize using AlphaEvolve's agentic harness to generate code; and Apply the optimized algorithm to production.Organizations are already seeing significant impact. BASF used AlphaEvolve to create a digital twin for their supply chain, improving planning and forecasting by over 80%. Coolblue optimized e-commerce demand forecasting, achieving a 5% reduction in WMAPE. FM Logistic improved warehouse routing by 10.4%, saving 15,000 km in staff travel.Infineon is using it for chip design, while JetBrains accelerated IDE performance by 15-20%. Kinaxis improved forecasting accuracy by over 22% and reduced runtime by 90%. Klarna doubled ML training pipeline throughput and improved model quality.Kuro Games saw substantial performance gains in server-side workloads. Oak Ridge National Laboratory deployed AlphaEvolve on the Frontier supercomputer to optimize GPU kernels. Old Dominion University used it to model biological aging mortality rates, rediscovering a known model and improving others.PacBio improved DNA sequencing accuracy by 30%. Pebble reduced GPU performance modeling errors by 56%. Qbraid used AlphaEvolve to refine quantum computing encodings.
https://cloud.google.com/blog/products/ai-machine-learning/alphaevolve-is-available-for-everyone/ cloud.google.com
CdXz5zHNQW_bOGGr5HEZK.png
RSS Hunter RSS Hunter • Yesterday

Autopilot Clusters with GKE managed DRANET: GPUs and TPUs

This blog post details setting up Google Kubernetes Engine (GKE) Autopilot clusters with managed DRANET to support GPUs and TPUs. GKE Autopilot simplifies Kubernetes management by handling nodes, scaling, and security, while managed DRANET enables requesting network resources for Pods, including those for TPUs and RDMA-enabled GPUs. The setup involves creating a Virtual Private Cloud (VPC), deploying an Autopilot cluster, and then configuring custom ComputeClasses and ResourceClaimTemplates.Configure essential variables such as region, cluster name, network, subnetwork, reservation URL, and Hugging Face token before proceeding. First, an Autopilot cluster is deployed, providing a fully managed environment. Next, a custom ComputeClass is created to specify the accelerator type (GPU or TPU), optionally referencing a reservation for resource allocation.Following this, a ResourceClaimTemplate is established, differentiating between RDMA-enabled GPUs (using deviceClassName: mrdma.google.com) and non-RDMA TPUs (using netdev.google.com). Finally, a workload is deployed, referencing both the custom ComputeClass and the ResourceClaimTemplate. This crucial step triggers GKE Autopilot to provision the specified node type and configure managed DRANET networking.The resource claim then acts as a bridge, directly binding the Pods to the accelerators on the provisioned nodes, ensuring the correct networking setup for the workload. This comprehensive process applies uniformly for both GPU and TPU deployments. Additional resources are provided for further exploration.
https://cloud.google.com/blog/topics/developers-practitioners/autopilot-clusters-with-gke-managed-dranet-gpus-and-tpus/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

C4N, now GA: Delivering cloud’s highest per vCPU network and block storage I/O for x86 workloads

Google Cloud's C4N instances are a new offering designed to address network and block storage performance bottlenecks in demanding enterprise applications. Built on Google's custom Titanium offload architecture, these instances offload network and storage tasks to dedicated hardware. C4N instances provide up to 400 Gbps of network bandwidth and 95 million packets per second (MPPS), offering significant improvements over comparable Intel-based offerings. This performance makes them ideal for network-intensive applications like virtual appliances, large-scale data analytics, and AI/ML workloads. When paired with Hyperdisk Extreme, C4N delivers Compute Engine's highest block storage performance, scaling up to 25 GiB/s and 1 million IOPS. The instances are powered by 5th Gen Intel Xeon Scalable processors and offer predictable, high-throughput I/O performance. C4N instances improve VM-to-VM and VM-to-internet network bandwidth and packet processing. They also offer optimized I/O for smaller VM shapes and faster Google Cloud Storage transfers without add-ons. C4N instances, when combined with Hyperdisk, allow for dynamic tuning of storage performance, latency, and throughput. Early customer adoption highlights the significant performance gains and TCO benefits of C4N for various data-intensive workloads.
https://cloud.google.com/blog/products/compute/c4n-network-and-storage-optimized-vms/ cloud.google.com
RSS Hunter RSS Hunter • Jul 8

New ways to keep Google Cloud certifications current and boost your career

Google Cloud certification is valuable, boosting recruiter responses, promotions, and salaries, with 81% of organizations trusting certified candidates more. However, rapid cloud technology changes mean skills have a short half-life, making recertification crucial. Google Cloud is simplifying this process with new technological approaches. They are moving beyond traditional, stressful, and time-consuming exams for recertification. You can now renew your Google Cloud credentials using Google Skills and skill badges. Google Skills allows you to select courses and badges relevant to your role and career goals. Taking updated courses can help refresh your knowledge on new topics. Alternatively, skill badges offer a faster path for those already applying new technologies at work through hands-on labs. This updated recertification process is available for Cloud Digital Leader, Associate Cloud Engineer, Professional Cloud Architect, and Professional Data Engineer certifications. Completing required activities while your certification is active automatically extends it by one year. This approach ensures your certification remains a living credential, demonstrating current readiness to employers.
https://cloud.google.com/blog/topics/training-certifications/new-ways-keep-google-cloud-certifications-current/ cloud.google.com
RSS Hunter RSS Hunter • Jul 8

Google Cloud named Leader in the 2026 Gartner® Magic Quadrant™ for AI Infrastructure

The agentic era of AI demands infrastructure that supports reasoning and action, not just responses. Google has been recognized as a leader in AI infrastructure by Gartner, highlighting their strong execution and vision. This success is built upon over a decade of co-designing silicon and software with Google DeepMind for services like Gemini. Google's custom TPUs are central to this, with TPU 8t optimizing training and TPU 8i enhancing inference for agents. They also partner with NVIDIA, offering A5X instances and integrating GPUs into Google Cloud services. Google actively contributes to open-source projects and offers TorchTPU for PyTorch developers.Their AI Hypercomputer offers improved performance per dollar through integrated hardware and software. Key components include high-bandwidth storage systems and the Virgo Network for massive compute connectivity. GKE Inference Gateway optimizes model scaling with low latency and reduced costs. This infrastructure allows fluid scaling, with Cluster Director and GKE enabling training up to 130,000 nodes. GKE Agent Sandbox provides secure, low-latency agent execution and cost optimization. Finally, Cross-Cloud Network and Cloud WAN ensure consistent performance across multi-cloud and hybrid environments.
https://cloud.google.com/blog/topics/ai-infrastructure/google-is-a-leader-in-gartner-magic-quadrant-for-ai-infra/ cloud.google.com
CdXz5zHNQW_ipuxr9KSlD.png
RSS Hunter RSS Hunter • Jul 8

Gemini Enterprise for Education named a Commander in Tambellini StarChart™: 2026 AI Agents for Administrative Efficiency—Agent Platforms

The agentic AI era is revolutionizing higher education, enhancing efficiency and innovation. The Tambellini Group has recognized Gemini Enterprise for Education as a "Commander," the highest category for AI agents focused on administrative efficiency, ranking it first in innovation and usability. This acknowledges Google's AI leadership and the capabilities of Gemini models and its agentic platform. Gemini Enterprise for Education uniquely consolidates Gemini models, agent-building tools, enterprise search, governance controls, and Google Cloud infrastructure. This integrated approach simplifies the development and management of AI services at scale for institutions. The platform offers a flexible, unifying foundation on Google Cloud, allowing for custom, unified architectures rather than siloed AI applications. Institutions can automate complex campus operations, shifting focus from manual tasks to strategic oversight, supporting use cases like advising and student services. Secure, governed AI agents empower students, faculty, and researchers with 24/7 adaptive support by grounding agents in institutional data through enterprise search and Deep Research capabilities. Features like administrative controls and role-based access ensure data protection and manageability as the technology scales. This recognition validates Google's significant investments in AI, security, and cloud technologies for education.
https://cloud.google.com/blog/topics/public-sector/gemini-enterprise-for-education-named-a-commander-in-tambellini-starchart-2026-ai-agents-for-administrative-efficiencyagent-platforms/ cloud.google.com
CdXz5zHNQW_Tjbujicdq7.png
RSS Hunter RSS Hunter • Jul 8

Meet the 33 cybersecurity startups joining the Gemini Startup Forum

Google has announced the selection of 33 startups for its Gemini Startup Forum, a program focused on AI-native cybersecurity. These startups are working on innovative solutions to tackle complex cybersecurity challenges, and they will have the opportunity to work with AI and cybersecurity specialists from Google DeepMind, Google Cloud, and Wiz. The startups are organized into six focus areas, including autonomous agent protection, application security, cloud and network security, endpoint security, SOC automation, and security infrastructure. Each startup is working on a unique solution, such as Capsule Security's runtime protection for autonomous AI agents or Aisy's vulnerability remediation platform. The Gemini Startup Forum is part of the Google for Startups program, which aims to support the next generation of AI-native cybersecurity startups. By collaborating with these startups, Google is committed to building a safer and more resilient digital infrastructure. The program provides startups with access to APIs, tools, training, and technical resources to help them scale with AI. The threat landscape is constantly evolving, and Google's commitment to supporting cybersecurity startups is crucial in meeting this challenge. The selected startups are well-positioned to make a significant impact in the cybersecurity industry, and their innovations will help define the next generation of proactive digital defense. Overall, the Gemini Startup Forum is an important initiative that brings together innovative startups and industry experts to tackle the most pressing cybersecurity challenges.
https://cloud.google.com/blog/products/identity-security/meet-the-33-cybersecurity-startups-joining-the-gemini-startup-forum/ cloud.google.com
RSS Hunter RSS Hunter • Jul 8

20 questions for the Agentic Enterprise (and how Agent Platform can help)

IT leaders are facing pressure to build and deploy AI agents quickly, but the underlying engineering complexity is significant. This complexity involves fragmented tools, data security concerns, and budget management. The Gemini Enterprise Agent Platform aims to simplify this by providing a unified environment for building, scaling, governing, and optimizing agents. To navigate these challenges, it's crucial to ask engineering teams specific questions.The build phase begins by understanding who is building the applications, as AI creation is no longer exclusive to high-code engineers. Developers need specialized AI tools to accelerate coding, but these often lack connection to essential enterprise data. Google Antigravity with specific extensions is recommended for core application, data, and Google Cloud engineers.It's essential to determine if agents are being built for human interaction or for other agents, as this dictates design requirements. For human interaction, focus on user experience; for agent-to-agent communication, prioritize interoperability using protocols like Agent2Agent. Choosing the right development tool involves considering a four-rung ladder: Agent Studio for low-code, Managed Agents API for agent-as-a-service, Antigravity 2.0 for advanced coding, and Agent Development Kit (ADK 2.0) for highly custom networks.For initial development, starting with a single, specialized agent is advised to maintain accuracy and efficiency. As complexity grows, transitioning to a multi-agent system where specialized agents collaborate is recommended. Connecting enterprise data requires open standards like Model Context Protocol (MCP) to provide agents with necessary context and logic for accurate decision-making.To ensure agents built on different frameworks can communicate, the Agent2Agent (A2A) protocol enables universal communication. Agents should dynamically retrieve needed tools using focused agentic Skills to avoid performance degradation and cost increases. Scaling requires deploying agents in a fully managed, serverless execution environment like Agent Runtime, offering elastic auto-scaling and secure private networking.To manage long-running tasks, agents need both short-term and long-term memory, with Agent Platform handling immediate session state and persistent storage. Limiting the blast radius for agents running scripts or browsing the web is crucial; this is achieved by executing such tasks in temporary, isolated sandbox environments.
https://cloud.google.com/blog/products/ai-machine-learning/20-questions-for-the-agentic-enterprise/ cloud.google.com
RSS Hunter RSS Hunter • Jul 7

Drive proactive security, prioritize risks with Google Threat Intelligence and Wiz ASM

Organizations are increasingly aiming to be more proactive in their security strategies. The rapid pace of AI-driven vulnerability discovery necessitates personalized security investments to prioritize defenses. Google Threat Intelligence is integrating with Wiz Attack Surface Management to address this need. This integration connects real-world exposures with real-time adversary activity. It aids in detecting and prioritizing external exploitable issues and uncovering logic-driven AI vulnerabilities. The goal is to enable a shift towards prioritizing actions based on the most significant real-world threats. Combining these insights moves security from reactive maintenance to a proactive stance. Organizations will gain visibility into which exposures are actively targeted by adversaries. The integration will automatically feed exposure data into Google Threat Intelligence for better prioritization. Wiz ASM maps external attack surfaces and scans for exploitable vulnerabilities and misconfigurations. The Wiz Red Agent uses AI to find complex logic-driven vulnerabilities, offering a more comprehensive view of risks.
https://cloud.google.com/blog/products/identity-security/drive-proactive-security-prioritize-risks-with-google-threat-intelligence-and-wiz-asm/ cloud.google.com
CdXz5zHNQW_fnvuXmj0hC.png
RSS Hunter RSS Hunter • Jul 7

A developer's guide to publishing agents in Gemini Enterprise and Google Cloud Marketplace

The landscape of software is shifting from Software-as-a-Service to Agents-as-a-Service, enabling AI agents to interoperate through standardized protocols like A2A. Developers can now build and commercialize these autonomous agents via Google Cloud Marketplace, integrating them with the Gemini Enterprise app. The process involves designing an architecture that bridges marketplace billing, security, and the Gemini Enterprise Agent Platform.Partners must join the Google Cloud Partner Network and adhere to specific listing requirements, including defining clear agent use cases and complying with the A2A protocol. A crucial element is the A2A Agent Card, a JSON file detailing the agent's capabilities, authentication methods, and endpoints. Agents must support public access or OAuth 2.0 for authentication and authorization.Technical requirements include following the A2A protocol documentation, which also guides the implementation of interactive user interfaces using A2UI. The A2A Agent Card is essential for displaying agent metadata and enabling the Gemini Enterprise app to locate endpoints, discover entry points, and determine authentication methods. Dynamic Client Registration (DCR) is supported to automate the OAuth client registration process.Publishing an agent involves selecting "AI Agent as a Service" in the Producer Portal, uploading the Agent Card, defining availability and pricing, and configuring backend procurement. Google Cloud then validates the agent before it becomes available on the Marketplace. The transaction and registration lifecycle involves distinct phases managed by a Billing Administrator, a Discovery Engine Administrator, and the end-user.Procurement is an asynchronous backend process initiated by a customer's subscription or private offer, triggering a notification to the partner. The partner approves the entitlement, records the transaction, and activates the subscription. Following procurement, the Discovery Engine Administrator links the purchase to their Gemini Enterprise environment through a synchronized registration flow, including a DCR handshake.Once registered, the agent becomes discoverable to end-users within the Gemini Enterprise app, who can then request access after proper identity authorization. This comprehensive process ensures secure, compliant, and governed access to third-party agents within an enterprise environment.
https://cloud.google.com/blog/topics/developers-practitioners/publish-agents-in-gemini-enterprise-and-google-cloud-marketplace/ cloud.google.com
CdXz5zHNQW_JOjJCHgP4L.png
RSS Hunter RSS Hunter • Jul 7

BGP route policies: Top 3 use cases by customer demand

Google Cloud's BGP route policies for Cloud Router empower network administrators with programmable control over network path evaluation and propagation. Policy named sets have been introduced to manage growing complexity by grouping prefixes or BGP communities into reusable entities, simplifying configurations. These policies, powered by CEL, allow fine-grained filtering and attribute modification for BGP routes within Cloud Router. Customers have leveraged these policies for route filtering and network protection, implementing "fail closed" models for enhanced security. They also use them to influence traffic paths for active/standby architectures by modifying MED attributes or AS-PATH prepending. A significant use case involves solving asymmetric routing for stateful firewalls by matching BGP communities. This allows Cloud Router to tag routes, adjusting MED for symmetrical return traffic. Customers are encouraged to test policies in staging environments before production deployment. Resources are available for implementing route filtering, influencing traffic paths, and creating traffic symmetry. Overall, BGP route policies offer a more robust and easier way to control dynamic routing for hybrid cloud connectivity.
https://cloud.google.com/blog/products/networking/bgp-route-policies-top-3-use-cases-by-customer-demand/ cloud.google.com
RSS Hunter RSS Hunter • Jul 7

Report: 83% of organizations need to upgrade their infrastructure to support agentic AI

The enterprise AI landscape has evolved from conversational bots to agentic AI that takes independent action. This shift strains existing infrastructure, with 83% of organizations needing upgrades for production-grade agentic AI. Legacy architectures struggle with the scale and continuous reasoning of autonomous agents, leading to an "inference tax" of data egress, storage bloat, and idle hardware. Fluid compute, matching silicon to tasks and minimizing overhead, is crucial for addressing these inefficiencies. Centralized governance is essential to manage the proliferation of autonomous agents and address security, governance, and MLOps challenges. A unified data layer allows agents to access and understand information regardless of its location, eliminating fragmentation. Hybrid multicloud architectures are becoming the norm, driven by digital sovereignty and the need to comply with local data residency laws. Edge AI deployment is critical for reducing latency, ensuring operational resilience, and improving cost-efficiency by processing interactions closer to the source. Energy consumption is a significant operational factor, influencing hardware selection due to grid scarcity, regulatory compliance, and infrastructure economics. Adopting unified, AI-optimized infrastructure, like Google Cloud's AI Hypercomputer, where all layers are co-designed, is key to overcoming these challenges. This holistic approach enables physical AI, where autonomous systems can interact with and solve real-world problems.
https://cloud.google.com/blog/products/compute/state-of-ai-infrastructure-report-overview/ cloud.google.com
CdXz5zHNQW_yqhRgZB4Qq.png
RSS Hunter RSS Hunter • Jul 7

The ‘Ghost’ in the Database: Recovering Active ADFS Signing Keys via Machine DPAPI

The "Golden SAML" technique allows attackers to forge identity assertions in Microsoft environments. By compromising an ADFS token-signing certificate's private key, an attacker can bypass security measures and impersonate any user. A new attack vector has emerged where configuration drift during manual certificate rotation can expose active signing keys in Machine DPAPI. This occurs when AutoCertificateRollover is disabled and the ADFS service uses a new certificate without updating the WID configuration database.This creates a "ghost" certificate entry that is no longer used for token signing. The active signing key, however, resides in the system's machine-scoped cryptographic store, protected by Machine DPAPI. Successfully retrieving this key enables attackers to forge valid SAML assertions, granting unauthorized access to SAML-federated applications. This method evades typical monitoring focused on LSASS and live ADFS processes.Attackers can exploit this by accessing the machine key store and related DPAPI artifacts. The private key is persisted in the machine-scoped key store, protected by Machine DPAPI for operational resilience. This resilience, however, means a privileged local process can recover the key material. The retrieved key can then be used to forge a SAML assertion, impersonating a high-privilege user like a Global Administrator.Defenders should monitor operating system cryptographic operations and identity issuance. SACL-based object access monitoring on specific file paths can provide supporting evidence. Inconsistencies in ADFS token issuance logs and federated identity monitoring in Entra ID are also crucial. Mitigation involves treating ADFS as Tier 0 infrastructure and considering hardware-backed key protection with Hardware Security Modules. Using gMSA for ADFS services can also reduce operational drift from manual credential management. Strict Tier 0 administrative controls are essential for ADFS servers.
https://cloud.google.com/blog/topics/threat-intelligence/recovering-active-adfs-signing-keys-machine-dpapi/ cloud.google.com
CdXz5zHNQW_pnUJepHCMM.png
RSS Hunter RSS Hunter • Jul 7

Google Cloud Labs: Accelerate AI with Cloud Run

The AI landscape is evolving beyond simple prototypes, requiring professional engineering for production-grade applications. Google Cloud is relaunching its Accelerate AI with Cloud Run roadshow for 2026, focusing on the full AI agent lifecycle. The "Coffee Shop Journey" is a hands-on experience demonstrating how to build and deploy advanced AI agents on Cloud Run for real business use cases. Participants will learn to deploy basic web apps, build recommendation agents with ADK and RAG, and optimize business locations using Gemma 4 and BigQuery. The curriculum also covers personal productivity assistants and advanced features with Antigravity 2.0, including skills, context, rules, and hooks. Cloud Run is presented as an ideal platform for scalable, production-grade agentic workloads. Hands-on exercises will cover building, scaling, and orchestrating long-running agents using Google's ADK and Antigravity 2.0. Attendees will also learn about data-driven expansion strategies with BigQuery MCP and low-latency inference on GPUs. The roadshow aims to empower participants to move from AI prototypes to powerful, production-ready realities. Workshops are scheduled in various North American cities throughout August, September, October, and November, with registration details provided for early events.
https://cloud.google.com/blog/topics/developers-practitioners/google-cloud-labs-accelerate-ai-with-cloud-run/ cloud.google.com
CdXz5zHNQW_M0EXIgiHEE.png
RSS Hunter RSS Hunter • Jul 7

Shift into high gear with agents: Securing the software-defined vehicle

The automotive industry is transitioning to software-defined vehicles (SDVs) that leverage over-the-air updates for new capabilities. Nexus SDV, a partnership between Google Cloud and Valtech, is a new AI-enabled connected vehicle platform designed to manage millions of devices. It integrates deeply with Android Automotive OS, streamlining data and in-vehicle user experiences. The platform's core showcases how Arm-based compute and Bigtable can reduce total cost of ownership. Nexus AI acts as the intelligent engine, using AI models to transform vehicles into proactive, agentic partners. This platform focuses on reducing operational costs by using efficient hardware and optimized data storage. Its cloud-native architecture ensures robust data flow between vehicles and data centers. Google Cloud's security framework provides defense-in-depth, protecting against threats and ensuring compliance. Nexus SDV offers an open-source framework for manufacturers to build the next generation of automotive intelligence securely and cost-effectively.
https://cloud.google.com/blog/products/identity-security/shift-into-high-gear-with-agents-securing-the-software-defined-vehicle/ cloud.google.com
CdXz5zHNQW_srNbXGBn5l.png
RSS Hunter RSS Hunter • Jul 6

Google’s Continued Disruption of Malicious Residential Proxy Networks

Google, in coordination with the FBI and others, has taken action against the NetNut residential proxy network. This action follows a previous disruption of the IPIDEA proxy network and is part of Google's ongoing effort to dismantle malicious residential proxy networks. Google disabled accounts and services used by NetNut for malware command and control. They also shared technical intelligence on NetNut's software development kits and infrastructure with platform providers and researchers. Google Play Protect will now warn users and disable applications incorporating NetNut SDKs, providing ongoing protection. These actions are believed to have significantly degraded NetNut's network operations, reducing the available devices by millions. NetNut operates by distributing SDKs to home devices, which then become part of the proxy network. These devices are used to route traffic, masking malicious activity by attackers. This poses risks to unsuspecting device owners, potentially flagging their legitimate traffic as suspicious. In June 2026, numerous threat clusters were observed using NetNut exit nodes for cybercrime and espionage. Consumers are advised to be cautious of applications offering payment for sharing internet bandwidth. Users should also stick to official app stores, review app permissions, and ensure security protections like Google Play Protect are active.
https://cloud.google.com/blog/topics/threat-intelligence/google-continued-disruption-residential-proxy-networks/ cloud.google.com
RSS Hunter RSS Hunter • Jul 2

SOCRadar powers rapid threat detection with AlloyDB and Gemini Enterprise

SOCRadar, a cybersecurity firm, faced performance issues with its PostgreSQL database due to increasing cyber threat volumes. They needed to modernize their infrastructure to provide faster threat intelligence to customers. Their existing on-premises database struggled with high-velocity data ingestion and real-time analytical queries, creating significant bottlenecks. To address this, SOCRadar evaluated alternatives and chose Google Cloud's AlloyDB for PostgreSQL.AlloyDB offered a PostgreSQL-compatible, fully managed solution capable of handling complex, hybrid workloads. This migration resulted in a 20x performance boost for analytical queries and a 3.2x increase in live ingestion velocity. The transition also significantly reduced operational overhead, freeing up database administrators. SOCRadar reclaimed over 45 TB of storage and saved 75% of their DBA time, enabling them to focus on innovation.Furthermore, SOCRadar integrated the Gemini Enterprise Agent Platform with AlloyDB to combat alert fatigue. This AI-driven filtering categorizes and routes alerts before they reach end-users, ensuring analysts receive only critical, actionable intelligence. The company is now exploring agentic AI workloads, with plans for natural language querying and automated incident summarization. This unified platform consolidates transactional velocity, historical depth, and AI intelligence for a future-proof cybersecurity defense framework.
https://cloud.google.com/blog/products/databases/socradar-powers-rapid-threat-detection-with-alloydb-and-gemini-enterprise/ cloud.google.com
RSS Hunter RSS Hunter • Jul 1

AlloyDB AI Functions - now with revolutionary performance boosts and cost savings

AlloyDB is an AI-native database that intelligently processes data, offering features like vector search and natural language-to-SQL capabilities. It integrates foundation models like Gemini directly through AI functions, which bring Gemini’s world knowledge to your data. These functions transform unstructured data, like raw user feedback, into structured insights, eliminating the need for complex custom pipelines. For example, ai.generate can convert raw text into clean, structured JSON.New AI functions include ai.summarize, ai.agg_summarize, and ai.analyze_sentiment, which categorize emotional tone, condense text, and summarize multiple rows, respectively. AlloyDB has achieved significant performance and cost improvements in AI function processing through two key breakthroughs. Smart Batching for AI Functions intelligently groups AI function calls, deduplicating prompt overhead and boosting performance up to 2,400 times. AlloyDB automatically determines optimal batch sizes and handles retries.Optimized AI Functions, initially for ai.if, deploy a small, proxy model trained on your data, reducing external LLM calls dramatically. This can process up to 100,000 rows per second and cut costs by 6,000 times. AlloyDB trains this proxy model in the background when you use the PREPARE statement and automatically falls back to the LLM if accuracy is low. These optimizations allow users to process complex queries, such as filtering products based on specific numerical constraints, with unprecedented speed and cost efficiency.Users can get started with AlloyDB’s 30-day free trial and enable AI functions directly within their SQL queries. To maximize performance and cost savings, users can leverage optimized functions and smart batching. These advancements empower users to bring Gemini’s intelligence to their AlloyDB data efficiently.
https://cloud.google.com/blog/products/databases/boost-performance-and-lower-costs-with-alloydb-ai-functions/ cloud.google.com
CdXz5zHNQW_WcdRjies8O.jpeg
RSS Hunter RSS Hunter • Jul 1

Get started with the Claude apps gateway for Google Cloud

Claude Code, an agentic coding tool from Anthropic, previously integrated with Google Cloud through individual developer configurations. However, scaling this across an organization presented challenges like credential management and lack of centralized control. The new Claude apps gateway addresses this by acting as a self-hosted service between local Claude Code clients and Google Cloud. This gateway centralizes identity, policy, cost management, and routing for improved governance. It integrates with identity providers for seamless user authentication without exposing sensitive credentials to developer machines. Role-based access control rules are managed centrally, overriding local settings and ensuring consistent policy enforcement. Telemetry data is attributed to verified users and groups, providing accurate usage metrics. The gateway also allows for setting granular spend limits per user or group, acting as a guardrail against excessive usage. Inference requests remain within the Google Cloud perimeter, maintaining data privacy and compliance. Setting up the gateway involves provisioning GCP resources, configuring the gateway.yaml file, and deploying it to Cloud Run. Developers are then onboarded through forced login settings, directing them to the gateway for authentication. This setup streamlines enterprise adoption and enhances security for Claude Code usage within Google Cloud.
https://cloud.google.com/blog/topics/developers-practitioners/announcing-claude-apps-gateway-for-google-cloud/ cloud.google.com
CdXz5zHNQW_27kEg3jWk2.png
RSS Hunter RSS Hunter • Jul 1

New IDC study: The business value of Mandiant Consulting

Security leaders now face the expectation of driving business growth and demonstrating security's value to their boards, beyond just risk management. Mandiant Consulting assists in translating technical security efforts into measurable financial benefits. A recent IDC study found that organizations partnering with Mandiant Consulting experienced an average annual benefit of $4.3 million, achieving a 268% ROI over three years with a quick 4.1-month payback period. These findings are based on interviews with large, complex organizations averaging $17.3 billion in revenue. Google Cloud views security as a strategic business enabler that directly impacts profitability. One healthcare client found that Mandiant improved their commercial conversations, positioning their security as a market differentiator and a key reason for client selection. Mandiant also helped reduce their insurance costs by $50,000 annually. Facing resource constraints, CISOs benefit from Mandiant's frontline threat intelligence, derived from extensive incident investigations, which helps teams focus on relevant industry-specific threats. A retail organization successfully defended against targeted campaigns like those from Scattered Spider, thanks to Mandiant's tailored detection use cases. Additionally, Mandiant conducts technical audits of identity infrastructure, providing leadership with reassurance and validation for their cybersecurity programs. This independent verification strengthens messaging to boards, as confirmed by an energy-sector organization. Overall, Mandiant customers report significant improvements in cyber-resilience, attack preparedness, and security analyst efficiency, allowing internal teams to focus on strategic growth initiatives.
https://cloud.google.com/blog/products/identity-security/new-idc-study-how-mandiant-transforms-security-into-a-competitive-advantage/ cloud.google.com
RSS Hunter RSS Hunter • Jul 1

Google Cloud confirmed to offer a safer choice for EU public sector organizations with Dutch DPIA approval

Google Cloud is dedicated to offering public sector organizations globally flexible, scalable cloud technology with top-tier data protection, sovereignty, and security. For European Union public sector entities, data protection is a fundamental requirement. A significant advancement in this commitment is the successful completion of a rigorous data protection impact assessment (DPIA) of Google Cloud by SLM Rijk, the Dutch government's strategic vendor management agency. This collaboration confirms Google Cloud's strong dedication to enhancing trust in its privacy practices within the Dutch public sector. All key concerns raised during the DPIA were successfully addressed, leading to the conclusion that no significant data protection risks exist when recommended measures are applied. Consequently, the Dutch central public sector can now officially utilize Google Cloud, with a clear privacy assessment pathway established. This outcome empowers Dutch public sector organizations and those beyond to confidently explore and adopt Google Cloud for secure modernization and digital transformation. This builds on the prior successful completion of a DPIA for Google Workspace in the Netherlands, reinforcing Google's commitment to privacy, security, and compliance for public sector clients. Google Cloud welcomes such independent assessments as vital for building trust and transparency. The company is committed to assisting customers in meeting their compliance obligations with secure and privacy-focused services. Google Cloud will continue to invest in privacy technologies, transparency, and customer controls to support organizations worldwide. They also offer resources to aid customers in navigating the complex DPIA process.
https://cloud.google.com/blog/products/identity-security/google-cloud-confirmed-to-offer-a-safer-choice-for-eu-public-sector-organizations-with-dutch-dpia-approval/ cloud.google.com
RSS Hunter RSS Hunter • Jul 1

Google named a Leader in 2026 Gartner® Magic Quadrant™ for Analytics and Business Intelligence Platforms for third year in a row

Google has been named a Leader in the 2026 Gartner Magic Quadrant for Analytics and Business Intelligence Platforms for the third consecutive year. This recognition follows Google Cloud Next 2026, where Google highlighted a shift towards a proactive system of action in data interaction. Looker and Google are bridging the gap between data insights and automated business workflows, powered by enterprise-grade trust and Gemini's reasoning. A universal semantic layer establishes a single source of truth, preventing data hallucinations and inconsistencies. Key governance strengths include unified analytics, in-database modeling, and enterprise lifecycle management with git-based version control. Gemini's capabilities enable complex strategic analysis for business users and accelerate analytics engineering for developers. Looker is reimagining its stack with agentic semantic modeling, dashboard agents, and conversational analytics. Looker Everywhere expands its footprint beyond traditional interfaces, integrating with external platforms. Specialized BI agents allow users to query complex data using natural language and orchestrate autonomous workflows. AI-powered self-service features in Explore Mode combine intuitive design with conversational analytics. Developers can now manage LookML development entirely within VS Code using the LookML Agent. Looker's semantic layer supports graph models and complex ontologies, enabling diverse data agent use cases. This integration ensures AI agents operate on verified enterprise metrics, not fabricated data.
https://cloud.google.com/blog/products/business-intelligence/looker-in-2026-gartner-analytics-and-bi-platforms-mq/ cloud.google.com
CdXz5zHNQW_Wtj5HMhrwT.jpeg
RSS Hunter RSS Hunter • Jul 1

Beyond Static Prompts: Building Scale-Proof, Polymorphic Multi-Agent Systems with Google's ADK

Enterprise generative AI faces scalability issues as it moves beyond simple chatbots to complex autonomous workflows. Traditional static prompting, where all schemas are pre-loaded, leads to context window bloat, high costs, and reduced accuracy due to attention diffusion. This architecture struggles with hundreds of data structures and dynamic business rules. A new approach is needed to decouple agent reasoning from structural data requirements. This post introduces Context-Aware Polymorphic Schema Validation, a pattern using a metadata registry for dynamic context injection and runtime schema enforcement.Static agent architectures lead to context window bloat, latency, and attention diffusion, where models mix irrelevant schemas. Maintaining synchronous code and validation becomes difficult, creating debt. Multi-agent handoffs lack deterministic checks, leading to silent failures. The proposed architecture externalizes schemas into a centralized metadata registry, separating execution into context discovery and dynamic validation. Schemas are stored as JSON descriptors containing field definitions, mapping rules, and validation hooks.The dynamic discovery and validation loop begins with a lightweight discovery prompt. The agent first distills user intent without heavy schema constraints. Once intent is clear, it loads specific schema rules from the metadata registry into session memory. The system then enters an evaluation loop, asking for precise fields and sending raw input to a separate Polymorphic Validator. If validation fails, an error code triggers self-correction; if it passes, the field is committed to the master JSON payload.Finalization occurs only when the master payload fully complies with metadata criteria, enabling secure downstream API calls or clean multi-agent handoffs. This design pattern, implemented on Google Cloud, utilizes ADK for multi-agent coordination and Gemini Flash for fast, cost-effective inference. Cloud Storage acts as the externalized storage for schema descriptors, which can be updated by administrators without code deployment. Cloud Run functions provide decoupled, programmatic validation hooks.This shift to dynamic schema architecture offers significant business and operational advantages. It ensures 100% reasoning density by avoiding context clutter, drastically reducing costs and hallucinations. Adaptability is achieved through zero-downtime updates by simply modifying schema descriptors in the registry. Deterministic state enforcement eliminates silent multi-agent failures by validating context before hitting enterprise applications.
https://cloud.google.com/blog/topics/developers-practitioners/beyond-static-prompts-with-google-adk/ cloud.google.com
CdXz5zHNQW_qpbSmsmj5P.png
RSS Hunter RSS Hunter • Jul 1

Scaling LLM Inference: Multi-Node KV Cache Offloading with GKE & Managed Lustre

Enterprise environments are adopting distributed architectures for AI workloads requiring long context windows. As these scale, KV caches exceed local CPU RAM and SSD limits. While pooling local SSDs offers capacity, it adds complexity in data distribution and replication. An alternative is to offload attention states to a high-performance external parallel filesystem like Google Cloud Managed Lustre. This approach bypasses host capacity limits and avoids networking overhead associated with pooled drives. Google Cloud Managed Lustre demonstrably saves TCO and reduces GPU-hour requirements for large language model inference. The solution can be extended with CPU RAM offload for further performance improvements. Key architectural components include GKE GPU nodes and Managed Lustre acting as a decentralized cache. A PVC Evictor manages storage by removing least-recently-used cache chunks. Deployment involves creating a GKE cluster, provisioning Lustre storage, deploying the vLLM serving engine with Lustre integration, and finally deploying the PVC Evictor. Proper Google Cloud project configuration, including quotas and IAM permissions, is essential before deployment.
https://cloud.google.com/blog/topics/developers-practitioners/scaling-llm-inference-multi-node-kv-cache-offloading-with-gke-managed-lustre/ cloud.google.com
CdXz5zHNQW_UPsefneDLL.png
RSS Hunter RSS Hunter • Jul 1

Conversational analytics in BigQuery brings trusted agentic reasoning to everyone

Conversational Analytics in BigQuery is now generally available, empowering both business and technical teams to analyze data using natural language. This feature provides an agent that acts like a knowledgeable analyst, built on Google's Gemini models and BigQuery’s secure foundation. It offers built-in conversational capabilities requiring no setup, with options for data professionals to create specialized agents grounded in specific data sources. These agents can access data beyond native BigQuery tables, including Lakehouse sources like Databricks, AWS Glue, SAP, and Salesforce, breaking down data silos.Data practitioners use Conversational Analytics within BigQuery Studio and Data Canvas, and can publish agents to Gemini Enterprise or other applications via an API. Engineered trust and explainability are key features, with every agent grounded in business context and providing visible thinking steps, SQL generation, and context citations. Proactive disambiguation through clarifying questions and long-term memory further enhance user experience and trust. Security and governance are inherited from BigQuery, ensuring users access only authorized data and all queries are logged for auditing.The product supports advanced security features like CMEK and VPC Service Controls, and guarantees data residency within EU and US multi-regions. Operational controls for scaling include cost management and usage tracking. Conversational Analytics leverages BigQuery’s AI functions, enabling users to ask questions about root causes, forecasts, and anomalies without building models. It can also query entire data estates, processing relational data and unstructured files like PDFs and images together.The agents are evolving from reactive analysis to proactive action with deep-dive mode, which automatically builds analytical plans for investigations. Agentic workflows allow for autonomous agents that monitor data, execute multi-step workflows on a schedule, and deliver insights directly. This release marks a shift from static dashboards to a self-managing environment that transforms data into active knowledge, forming a key component of the Agentic Data Cloud.
https://cloud.google.com/blog/products/data-analytics/conversational-analytics-in-bigquery-now-ga/ cloud.google.com
CdXz5zHNQW_2DWjoM5ccQ.gif
RSS Hunter RSS Hunter • Jun 30

How Schrödinger sped up molecular discovery by 4x with Alphaevolve

Computational chemistry traditionally faced a speed-accuracy trade-off in molecular simulations. Machine-learned force fields (MLFFs) improved accuracy but still struggled with the massive datasets needed for modern drug discovery. To address this, Schrödinger partnered with Google Cloud and used AlphaEvolve, an AI coding agent, to optimize their MLFF training pipeline. They identified two key algorithms, neighbor list computation and Ewald summation, as performance bottlenecks. The primary goal was to accelerate AI model training, specifically targeting the computationally demanding Ewald summation in their PyTorch code. AlphaEvolve was tasked with generating a more efficient implementation of this algorithm. The system successfully evolved the PyTorch code by replacing slow for-loops with parallel batch matrix multiplication for the Ewald summation. Rigorous evaluation confirmed the evolved code's functional correctness and performance improvements, with a success rate increasing significantly. This optimization resulted in a four-fold speedup in MLFF training and inference, accelerating drug discovery, catalyst design, and materials development. Schrödinger plans to further explore this evolutionary approach for optimizing custom GPU kernels.
https://cloud.google.com/blog/products/ai-machine-learning/schrodinger-alphaevolve-molecular-discovery-accelerates-4x/ cloud.google.com
RSS Hunter RSS Hunter • Jun 30

Modernizing financial services with deployment freedom and transformational AI with AlloyDB Omni

The financial services industry faces immense pressure from regulatory compliance, extreme speed demands, and robust security needs. Historically, proprietary databases have led to technical debt and high operational costs. Financial institutions struggle with licensing traps, where maintenance costs escalate, and innovation is stifled. Emerging regulations like DORA and data residency laws challenge public cloud adoption for sensitive data. This creates an "insights gap" as legacy systems cannot handle real-time data processing and AI-driven tools. The advent of Agentic AI necessitates a new database strategy moving beyond proprietary systems. A hybrid, open-standards approach is needed, combining cloud-native innovation with on-premises data control. Google Cloud's AlloyDB for PostgreSQL offers a solution, providing PostgreSQL compatibility and flexible deployment options. AlloyDB Omni specifically addresses these challenges by enabling on-premises, edge, or hybrid cloud deployments with superior performance and scalability. It offers modernized infrastructure, operational simplicity through containerization or standalone versions, and significantly improved transactional and analytical speeds. Furthermore, it integrates AI capabilities and enterprise-grade security features to meet stringent financial industry standards. This allows financial institutions to achieve both stability and innovation without being forced into public cloud migrations.
https://cloud.google.com/blog/products/databases/alloydb-omni-secure-hybrid-database-modernization-for-finance/ cloud.google.com
RSS Hunter RSS Hunter • Jun 30

Bringing speed and strong cost performance to the market with Gemini Omni Flash and Nano Banana 2 Lite

Google is introducing two new models to enhance creative workflows and reduce generation times. Nano Banana 2 Lite is now generally available, offering the fastest and most cost-effective image generation and editing within its family. This model allows for rapid idea exploration, A/B testing, and scaling for applications. Gemini Omni Flash is also being released in public preview, excelling in high-quality video generation and conversational editing. This model provides precise control for tasks like character swaps and scene relighting. Both models offer competitive price-performance for image and video creation. Gemini Omni Flash supports multimodal inputs and integrates world knowledge for realistic storytelling. It also synchronizes text and action within videos. Nano Banana 2 Lite boasts significantly improved visual quality and capabilities over its predecessor. It enables quick drafting of contextual scenes and maintains character consistency. Both models incorporate C2PA content credentials and SynthID watermarking for authenticity. Businesses like Adobe, Invideo, and WPP are already leveraging these new Gemini models. Developers can begin integrating these advanced image and video generation and editing features into their applications.
https://cloud.google.com/blog/products/ai-machine-learning/nano-banana-2-lite-and-gemini-omni-flash-available/ cloud.google.com
CdXz5zHNQW_GPG413iMjf.png
RSS Hunter RSS Hunter • Jun 30

Build agents even faster with Gemini Enterprise Agent Platform’s fully-managed, remote MCP server

Google Cloud now offers a Gemini Enterprise Agent Platform remote MCP server. This server acts as a secure bridge between external AI agents and Google Cloud resources. Developers can connect their Integrated Development Environments (IDEs) to Google Cloud, enabling AI agents to interact with services like Model Garden and Notebooks. This integration simplifies development by reducing the need for extensive integration code. It also accelerates time-to-value by providing a standardized interface within Google Cloud's secure infrastructure. The platform supports open standards, preventing vendor lock-in and ensuring compliance. Agent Registry provides a centralized library for discovering and governing AI capabilities. Security is maintained through native Cloud IAM Deny policies, controlling resource access. Enabling the Gemini Enterprise Agent Platform API automatically activates the remote MCP server. Connecting involves configuring the AI application to point to the remote server and utilizing available toolset endpoints. These toolsets facilitate interaction with generative AI, prediction, notebook management, model endpoints, model registry, fine-tuning, evaluation, and prompt management. Developers can start building by connecting their agent frameworks to the Agent Platform MCP server.
https://cloud.google.com/blog/products/ai-machine-learning/gemini-enterprise-agent-platform-remote-mcp-server/ cloud.google.com
CdXz5zHNQW_GH6gfCdQ16.png
RSS Hunter RSS Hunter • Jun 30

Anomaly detection using dynamic thresholds and two-year-long alerts in Cloud Monitoring

Setting alert thresholds has been challenging due to the need for historical data analysis and the limitations of static thresholds with growing workloads. Metrics that vary by time of day are particularly difficult to monitor with fixed thresholds. Cloud Monitoring alerts now offer long lookback alert policies for PromQL, currently in preview, which utilize up to two years of metric data. This feature enables dynamic thresholding, where alerts trigger based on a metric's historical behavior rather than a fixed value. For instance, an alert can be set to fire if recent performance deviates significantly from a longer historical average. Various algorithms are available for dynamic thresholding, including moving averages for stable data, z-scores for volatile data, and seasonal decomposition for time-of-day patterned metrics. Moving averages compare recent trends to a long-term average, while z-scores identify anomalies based on standard deviations. Seasonal decomposition compares current data to corresponding historical periods, ideal for metrics with daily or weekly patterns. Dynamic thresholds can also be used to prevent cost overruns by automatically adjusting quotas when spend metrics show anomalous increases. Cloud Monitoring is developing more advanced anomaly detection features using AI models. Users can sign up as design partners to provide feedback on these new capabilities.
https://cloud.google.com/blog/products/management-tools/cloud-monitoring-adds-long-lookback-alert-policies-for-promql/ cloud.google.com
CdXz5zHNQW_9xiSqFQTjK.png
RSS Hunter RSS Hunter • Jun 30

Supercharging the agentic era with Spanner’s multi-model architecture

The agentic era demands databases to be active context engines for AI, not just passive storage. Spanner is presented as a leading solution, offering a unified, multi-model foundation essential for generative AI and autonomous workflows. Gartner has recognized Spanner's efficiency, ranking it #1 in Lightweight Transactions for operational cloud database management systems. A Forrester study highlighted significant economic value, showing a 132% ROI and a fast payback period for Spanner deployments. True AI autonomy requires deep context, which Spanner provides by natively integrating relational, vector, graph, key-value, and full-text search data. This multi-model integration allows AI to access situational, semantic, and relationship context concurrently. Spanner features include unified graph and relational experiences, integrated vector search, high-performance relational and key-value capabilities, and advanced full-text search. An integrated columnar engine speeds up analytical queries on live data, eliminating the need for extensive ETL processes. Spanner's interoperability allows complex queries combining different data models in a single SQL statement. Spanner Omni extends these multi-model capabilities to any environment, including on-premises and other cloud providers, without hardware restrictions. The platform's underlying architecture leverages technologies like TrueTime and Paxos for global consistency and integrates advanced features like ScaNN-powered vector search and dynamic resharding. Spanner is positioned as a foundational element of Google's Agentic Data Cloud, aiming to break down data silos and enable autonomous AI agents.
https://cloud.google.com/blog/products/databases/the-power-of-multi-model-spanner-for-the-agentic-era/ cloud.google.com
CdXz5zHNQW_rAOxobwcjz.png
RSS Hunter RSS Hunter • Jun 29

Cloud CISO Perspectives: How Google Cloud Security uses AI internally

Google Cloud is transitioning to autonomous software development lifecycle security using AI agents. These agents are embedded across all stages of development, from design to production, to continuously strengthen products. Manual processes like threat modeling are replaced by agent-based reviews that cross-reference designs against security requirements. Google Cloud developed the Mantis framework for scalable, context-aware AI code scanning, which condenses repositories to reduce overhead. Specialized agents within Mantis analyze code structure, investigate files, and filter findings to eliminate false positives. A reproduction sandbox verifies exploitability before alerting developers. Self-healing fuzz testing is employed to uncover runtime vulnerabilities without manual intervention. An autonomous patching pipeline automates the reproduction, context mapping, fix generation, and validation of vulnerabilities. Post-launch, an autonomous security posture management system continuously monitors production systems for configuration drift. Through continuous augmentation via self-reflection, successful security patterns are stored to improve future agent performance, moving towards "immune" software that self-heals.
AI and ML News on Bluesky @ai-news.at.thenote.app bsky.app
CdXz5zHNQW_yD5A560UnL.png
RSS Hunter RSS Hunter • Jun 29

Synthesize the big picture and analyze trends with BigQuery's AI.AGG function

BigQuery has introduced AI.AGG(), a new function allowing natural-language instructions in SQL to summarize or synthesize millions of unstructured data rows. This function is invaluable for analyzing data where individual row analysis is insufficient, such as system logs or product reviews. AI.AGG() automates the process of extracting key insights from large volumes of text and even multimodal data. For instance, it can identify top feature requests from negative reviews or common user errors from system logs. The function efficiently handles large datasets by batching and aggregating data, simplifying context window management for users. It also supports both text and image inputs, enabling analysis of visual data like product photos. To ensure comprehensive analysis, it's best practice to pre-filter data and reduce input tokens before using AI.AGG(). Users can specify desired AI models for query execution, controlling the AI endpoint used. NULL values in input rows are automatically skipped, requiring careful handling of structured data to avoid silent data loss. AI.AGG() aims to provide partial results even with errors, and users can review job statistics for processing failures. This powerful function is currently in preview and available for all BigQuery users to explore.
AI and ML News on Bluesky @ai-news.at.thenote.app bsky.app
CdXz5zHNQW_tm0jnqClJO.png
RSS Hunter RSS Hunter • Jun 29

Scaling Network Analysis for Fraud Prevention with BigQuery Graph

Curve, a UK-based financial super-app, faces challenges with high-volume financial crime. Traditional fraud detection struggles with organized fraud rings that exhibit complex interconnections. To address this, Curve partnered with Google Cloud to implement BigQuery Graph for deep network analysis. Identifying fraud requires understanding multi-hop relationships, which are computationally expensive and difficult to scale with standard SQL. Curve transitioned its network analysis to BigQuery Graph, leveraging its native Graph Query Language (GQL) support. This allowed them to keep data within their existing BigQuery environment, saving time and costs. By modeling their payment ecosystem as a property graph, Curve simplified architecture and uses intuitive GQL for pattern matching. They can now traverse billions of connections efficiently and unify data analysis with standard SQL and machine learning workflows. This integration has led to significant financial impact, with estimated savings of $12 million in transaction losses for 2025. Graph-powered queries achieve approximately 72% accuracy in identifying fraudulent users, allowing fraud agents to focus on high-certainty cases. The adoption of GQL has streamlined fraud rules and enabled faster traversal for machine learning models. Curve is now exploring real-time detection loops and native graph visualization to further enhance its fraud mitigation strategy.
https://cloud.google.com/blog/products/data-analytics/fraud-prevention-with-bigquery-graph/ cloud.google.com
RSS Hunter RSS Hunter • Jun 29

The Bear Necessities: A Look at the Drivers, Dynamics, and Applications of the Pro-Russia Influence Ecosystem

Russia's pro-influence ecosystem, initially focused on supporting the Ukraine war, has evolved into a global strategic asset. This evolution includes expedited development of new influence assets and a resurgence of pro-Russia hacktivism. The shift indicates a growing focus beyond Ukraine, potentially intensifying influence operations targeting the EU and NATO. The war has served as a critical feedback loop, refining Russia's influence activities for global application. Increasingly, generative AI is being used for planning, research, and content creation in these operations. New actors are adopting influence tactics, signaling Russia's reliance on these cost-effective methods. The interconnected nature of this ecosystem makes it resilient to disruptions. Russia's influence strategy, rooted in Soviet-era "active measures," now blends overt and covert digital operations. Core objectives include diminishing Western primacy, retaining regional dominance, and ensuring domestic political stability. Targeting has pivoted from Ukraine to a broader scope including the US, Europe, the "Near Abroad," the Middle East, Africa, and domestic Russia. Tactics like narrative resonance, cyber-enabled influence, media mimicry, and direct dissemination are employed. The ecosystem comprises diverse components, from official channels to deniable covert actions, operating in a self-sustaining manner. This interconnectedness enhances the utility of Russian influence operations.
https://cloud.google.com/blog/topics/threat-intelligence/pro-russia-influence-ecosystem/ cloud.google.com
CdXz5zHNQW_Rnf1NjGOnD.png
RSS Hunter RSS Hunter • Jun 29

Turner Industries’ Blueprint for a Secure, Cloud-First Infrastructure

Turner Industries, an industrial contractor, sought to enhance efficiency and reduce costs. They implemented a new technology stack comprising ChromeOS, Google Workspace, Chrome Enterprise Premium, and Cameyo. This integration aimed to provide a secure and efficient foundation for their operations. The adoption of Chromebooks resulted in a 40-50% cost reduction per device and doubled device longevity. This led to significant savings, amounting to $700,000 across 1,200 devices. Furthermore, device configuration and deployment time was drastically reduced, saving an estimated 1,750 hours annually. Google Workspace and Chrome Enterprise Premium were adopted for all 21,000 employees. ChromeOS also streamlined IT support by minimizing time spent on patching and security tickets. Chrome Enterprise Premium enabled centralized endpoint management and security policy enforcement. Cameyo provided a solution for accessing legacy Windows applications, simplifying their VDI architecture. The new system is now used across all departments, including leadership, demonstrating its versatility. This technological transformation has led to a simpler, more secure, and efficient infrastructure for Turner Industries.
https://cloud.google.com/blog/products/chrome-enterprise/turner-industries-blueprint-for-a-secure-cloud-first-infrastructure/ cloud.google.com
RSS Hunter RSS Hunter • Jun 29

Securing agentic AI with perimeter guardrails: What's new in VPC Service Controls

Autonomous AI agents require robust architectural guardrails for safe innovation as they scale. To protect data, clear network-level boundaries are essential for these agents that connect across tools and datasets. Google Cloud recommends VPC Service Controls (VPC-SC) to establish a crucial destination-based perimeter for these workflows. New capabilities in VPC-SC enhance AI security by strengthening these boundaries.Agent identities can now be directly added to service perimeter rules, allowing for least-privilege access and immediate revocation if an agent is compromised. Conditional access rules can be enforced using model context protocol (MCP) attributes, enabling granular control over tool interactions, like denying email sending while allowing read access. VPC Service Controls are now natively integrated with the Gemini Enterprise Agent Platform, automatically blocking public internet access for enhanced security.This integration secures agentic workloads by reinforcing network-level security controls, acting as a foundational layer for data protection. VPC-SC, alongside identity and resource controls, forms a layered approach to enterprise AI security. It defends against unique attack vectors, acting as a critical network safety net against compromised agents. VPC-SC effectively prevents data exfiltration even when agents have valid IAM credentials.
https://cloud.google.com/blog/products/identity-security/securing-agentic-ai-whats-new-in-vpc-service-controls/ cloud.google.com
CdXz5zHNQW_67wexxbhIr.jpeg
RSS Hunter RSS Hunter • Jun 26

From query to action: Introducing SQL alerting in Cloud Monitoring Observability Analytics

Traditional alerting systems struggle with high-cardinality data and complex relationships, often forcing a compromise between immediate, noisy alerts and rigid metric monitoring. Critical system issues are frequently hidden within aggregated data and signal correlations. Observability Analytics now allows users to query logs and traces using SQL, and importantly, to create alerts from these complex analytical queries. This SQL alerting capability in preview moves beyond basic threshold monitoring to deep, contextual detection.Alerting policies run scheduled SQL queries, analyzing recent data based on a lookback window. If query results meet a defined condition, a notification is sent through configured channels like email or Slack. The system leverages BigQuery for processing, incurring associated costs. Two alert trigger types are available: a row count threshold for simple event volume monitoring, and a boolean condition for more sophisticated logic directly within SQL queries.For example, an e-commerce operator can detect payment gateway outages by alerting on a spike in gateway timeout errors using a row count threshold. An AI platform engineer can monitor agent latency by querying trace data and alerting if the 99th percentile latency exceeds a specified limit using a boolean condition. Setting up SQL-based alerts requires enabling Observability Analytics for logs or traces, linking a BigQuery dataset, and configuring necessary IAM permissions and notification channels. Creating an alert involves composing and validating an SQL query in Observability Analytics, selecting the BigQuery engine, defining the alert condition and schedule, and configuring notification channels. Alerts can also be managed via API and Terraform for Infrastructure as Code.
https://cloud.google.com/blog/products/management-tools/alert-with-sql-in-cloud-monitoring-observability-analytics/ cloud.google.com
CdXz5zHNQW_5AyOYt4eWl.png
RSS Hunter RSS Hunter • Jun 26

STOCKSTAY Another Day: The Latest Addition to Turla’s Intelligence Gathering Apparatus

Google Threat Intelligence Group has detailed STOCKSTAY, a .NET backdoor actively developed by the Russia-linked Turla group since late 2022. Turla has deployed STOCKSTAY primarily against Ukrainian government and military entities, as well as organizations interested in Italian foreign policy, for cyber espionage. This backdoor exhibits significant code and functionality overlaps with Turla's previously known KAZUAR toolkit. Turla is a long-standing cyber espionage actor, with suspected activity dating back to 2004, and has been linked to Russia's FSB. STOCKSTAY is a multi-component backdoor that communicates with its command and control (C2) server using secure WebSockets, facilitated by the websocket-sharp library. Initially designed to mimic a stock market data viewer, STOCKSTAY has evolved to impersonate other benign applications like PDF viewers and calculators. The malware comprises distinct components: STOCKSTAY.STOCKBROKER for network communication, STOCKSTAY.STOCKMARKET as the orchestrator responsible for configuration and C2 messaging, and STOCKSTAY.STOCKTRADER for executing commands on the infected host. STOCKSTAY.STOCKTRADER supports a range of operations including file and registry manipulation, command execution, and system information gathering. A related downloader component, STOCKSTAY.MARKETMAKER, has been observed masquerading as legitimate software to establish persistence. The analysis provides a timeline of observations and examines STOCKSTAY's similarities to KAZUAR, contextualizing its role within Turla's evolving toolset.
https://cloud.google.com/blog/topics/threat-intelligence/stockstay-turla-intelligence-gathering/ cloud.google.com
CdXz5zHNQW_ezln4WcZu8.png
RSS Hunter RSS Hunter • Jun 25

Enhanced data resilience with cross-region backups in Backup and DR Service

Organizations need a strong data backup strategy for business continuity. Multi-region backups offer high availability but can be expensive. A new, more cost-effective solution called cross-region backups is now generally available. This allows backup destinations to be in a different region than the source data. This decoupling protects against regional outages while respecting data residency rules. Currently, it supports Compute Engine instances, Disks, and Filestore, with Cloud SQL and AlloyDB to follow. Cross-region backups optimize costs by allowing granular selection of recovery regions. They also simplify compliance with data residency laws like GDPR. This feature improves regional resilience by storing data copies in entirely different geographical areas. Implementing cross-region backups involves creating a backup vault in a separate region. Then, a backup plan is configured in the resource's region but points to the secondary vault. Finally, attaching the plan automates data transfer to the regional backup vault. This new capability provides resilience and cost-efficiency without sacrificing compliance.
https://cloud.google.com/blog/products/storage-data-transfer/backup-and-dr-service-adds-cross-region-backups/ cloud.google.com
CdXz5zHNQW_2NGzetp1sa.png
RSS Hunter RSS Hunter • Jun 24

Zero-Day Exploitation of Vulnerability (CVE-2026-20245) in Cisco Catalyst SD-WAN Manager

In early 2026, Mandiant detected a threat actor targeting a service provider's SD-WAN infrastructure. The actor gained initial access through rogue peering connections, manipulating default account passwords. They then exploited a zero-day vulnerability, CVE-2026-20245, in Cisco Catalyst SD-WAN. This allowed them to escalate privileges from an administrative account to root-level access. The vulnerability exploited a flaw in the device's file upload feature which lacked proper filtering of malicious data. The threat actor also utilized anti-forensic techniques, selectively deleting and restoring system configuration files. They even executed a validation script to ensure all traces of their activities were purged. This campaign highlights the growing trend of compromising network appliances to bypass traditional security perimeters. SD-WAN orchestrators are becoming prime targets due to their central control and limited telemetry. Organizations are urged to patch Cisco Catalyst SD-WAN Manager immediately and follow hardening guidelines for enhanced security.
https://cloud.google.com/blog/topics/threat-intelligence/zero-day-exploitation-cisco-catalyst-sd-wan-manager/ cloud.google.com
RSS Hunter RSS Hunter • Jun 24

Log Analytics is now Observability Analytics: Query logs and traces with SQL

Google Cloud has enhanced its Observability suite with new capabilities for developers and SREs. Observability Analytics, formerly Log Analytics, now integrates trace data generally available to provide a unified experience. This allows for detailed root-cause analysis of traditional and agentic workloads. The Observability API for management and configuration is also generally available. A key feature is support for SQL in Cloud Trace. This enables joining application logs with trace spans for in-depth analysis. For AI agents, it allows analysis of telemetry across thousands of runs to identify tool failures and performance bottlenecks. Observability Analytics leverages BigQuery and SQL for direct analysis of telemetry data. This unification breaks down data silos, accelerating troubleshooting and enabling business correlation by joining observability data with business metrics. In-place analysis reduces data duplication and storage costs. Use cases include optimizing AI agents by analyzing tool failures and latency at scale. It also helps identify specific customers experiencing performance issues. While log and trace explorers are for individual investigations, Observability Analytics is designed for aggregations and broad analytical questions. The Observability API also allows AI agents to query telemetry programmatically through BigQuery. Users can start analyzing trace data in Observability Analytics today via the Google Cloud console.
https://cloud.google.com/blog/products/management-tools/query-logs-and-traces-with-sql-in-observability-analytics/ cloud.google.com
RSS Hunter RSS Hunter • Jun 23

Verifiable, private AI: Google Cloud expands Confidential Computing frontiers

Google Cloud announces significant innovations in Confidential Computing to enhance data privacy for AI deployments. Confidential Computing uses hardware-based Trusted Execution Environments (TEEs) to cryptographically protect data while it's being processed. Global scale Confidential AI capabilities now allow AI inference and fine-tuning with enforceable privacy guarantees. New Confidential G4 VMs with NVIDIA RTX PRO 6000 Blackwell GPUs are available globally, offering accessible Confidential AI. These VMs, powered by AMD EPYC CPUs and AMD SEV, protect data during processing and encrypt data transfer between CPU and GPU. Open-source Prompt Encryption SDKs provide end-to-end cryptographic protection for AI prompts and responses. Google Cloud is also collaborating with Apple to extend Apple's Private Cloud Compute on its platform, leveraging Confidential Computing and Intel TDX. Intel TDX is coming soon to C4 machine series Confidential VMs, providing hardware-isolated Trust Domains. Live Migration on C3D-based Confidential VMs is now generally available, enabling maintenance without workload interruption. Confidential Space, designed for secure multi-party computation, now integrates with Intel Trust Authority for independent verification. Additionally, Confidential Space now supports NVIDIA Hopper GPUs for secure, multi-party AI and machine learning workloads. These advancements aim to make Confidential Computing a foundational layer for secure collaboration and private AI innovation.
https://cloud.google.com/blog/products/identity-security/verifiable-trust-in-the-ai-era-whats-new-in-confidential-computing/ cloud.google.com
CdXz5zHNQW_5ljCN2g63H.png
RSS Hunter RSS Hunter • Jun 23

Open models, global networks: How AT&T and GSMA are accelerating telecom innovation with Gemma

Telecommunications is a complex field with proprietary data, making general AI models struggle to understand network operations. The industry requires extreme precision due to its massive scale, yet only a small percentage of AI is deployed on networks because of specialized knowledge gaps. Domain-specific AI models are essential for achieving the automation and workflows needed in the AI era, and open models offer the required flexibility. General frontier models lack the foundational context and understanding of specialized telecom vocabulary, topologies, and vendor data. Telco-specific models, trained on relevant datasets, can accurately interpret technical logs and diagnose network issues. The GSMA launched the Open Telco AI platform to build trusted, telco-grade AI, using Google's Gemma models as a foundation for AT&T's OTel models. These models were trained on specialized telco datasets curated by industry collaborators, resulting in numerous optimized models built with safety and reduced hallucinations in mind. AT&T's testing revealed that Gemma models performed exceptionally well after telecom-specific fine-tuning, with one Gemma model achieving high accuracy. These open telco models have seen significant downloads and outperform larger general models on specific benchmarks. Google Cloud is committed to supporting telecom operators in developing and deploying custom telco AI models, enabling accelerated AI adoption and safe deployment.
https://cloud.google.com/blog/topics/telecommunications/open-models-global-networks-how-att-and-gsma-are-accelerating-innovation-with-gemma/ cloud.google.com
RSS Hunter RSS Hunter • Jun 23

Boost BigQuery with Python: Managed Python UDFs now generally available

SQL is excellent for structured data analysis, but struggles with complex procedural logic, scientific computations, and machine learning. Python excels at these tasks. Previously, data practitioners faced infrastructure management overhead to run custom Python code. BigQuery Managed Python User-Defined Functions (UDFs) are now generally available, addressing this challenge. This feature allows users to execute custom Python code directly within BigQuery using SQL or BigQuery DataFrames. These UDFs leverage fully managed serverless resources that automatically scale. They offer flexibility by enabling access to the vast Python ecosystem, including libraries like NumPy and scikit-learn. Furthermore, they allow real-time integration with external APIs and Google Cloud services. Advanced users can optimize performance through vectorized processing with Pandas PyArrow, configurable container resources, and customizable concurrency. Streaming logs and real-time metrics facilitate debugging and monitoring. Python UDFs are billed under the BigQuery Services SKU and are eligible for cost commitments. Getting started involves exploring product documentation and public datasets.
https://cloud.google.com/blog/products/data-analytics/python-udf-in-bigquery-now-generally-available/ cloud.google.com
RSS Hunter RSS Hunter • Jun 22

The Starter Tier for Google AI Studio explained

Google AI Studio now offers a Starter Tier to quickly deploy prototypes with a live URL. This tier provides a pre-wired stack of Google Cloud services like Cloud Run, Cloud Firestore, Cloud SQL for PostgreSQL Developer edition, and Firebase Authentication. Google provisions and manages these resources in a fully-managed project behind the scenes, eliminating the need for a payment method or billing account initially. Individual Google Accounts can use this tier, with potential restrictions for Workspace users. The Starter Tier offers a simplified console experience focused on essential prototype needs. It is governed by separate Starter Tier Additional Terms, not the standard Google Cloud Terms of Service. Cloud Run handles compute and scales automatically, allowing up to two active web applications. Firebase Authentication with Google Sign-In is included for user login, simplifying integration for Google Workspace apps. Cloud Firestore provides NoSQL data storage, with AI-generated security rules for initial setup. Cloud SQL for PostgreSQL Developer edition is available for relational data needs. Deployment from prompt to live URL is a five-step process involving describing the app, enabling Firebase if needed, publishing, and obtaining the URL. The Starter Tier has limits, including a two-app cap, a single region, locked APIs, and shared Firestore quotas. These limits are generous for prototyping but necessitate an upgrade for more extensive needs. Upgrading to a paid account involves setting up billing and offers full Google Cloud platform access with more control and scalability. Budget alerts and instance caps are recommended to manage costs after upgrading.
https://cloud.google.com/blog/topics/developers-practitioners/the-starter-tier-for-google-ai-studio-explained/ cloud.google.com
CdXz5zHNQW_J7SImUZ6pB.png
RSS Hunter RSS Hunter • Jun 22

Scaling Ray Serve LLM on GKE: Performance without losing the developer experience

Ray Serve, a scalable model serving library, combined with Google Kubernetes Engine (GKE) offers a powerful platform for LLM serving. Historically, Ray Serve's flexibility came at a performance cost. However, through a partnership with Anyscale, Ray Serve now offers significantly improved performance, achieving up to 5x higher throughput and 8x lower latency. These advancements are due to three key architectural optimizations. Ray Serve now integrates HAProxy for efficient request routing and load balancing, reducing proxy overhead. A direct token streaming architecture bypasses the ingress router for token streams, cutting latency. The v2 Ray executor backend for vLLM enables asynchronous scheduling, unifying the code path and closing the performance gap. Benchmarks on GKE with next-generation AI hardware demonstrated these dramatic performance improvements. The enhanced Ray Serve scales throughput while maintaining low latency, even with increasing concurrent users. GKE provides the necessary infrastructure for these optimizations, offering automated scaling, monitoring, and fault tolerance. Developers can now achieve production-grade performance on Kubernetes without sacrificing Ray's rich features. The latest Ray release (2.56 and later) incorporates these enhancements, and resources are available for further exploration.
https://cloud.google.com/blog/products/containers-kubernetes/improving-ray-serve-llm-on-gke-throughput-latency/ cloud.google.com
CdXz5zHNQW_9uXpF6z40X.png
RSS Hunter RSS Hunter • Jun 18

Scaling the Next Generation of Global Innovation: How Google Supports Top Startups Around the World

The transition from a tech prototype to a successful business requires more than funding, demanding architectural guidance and scalable systems.  Google's accelerator program supports this by connecting startups with Google's engineering expertise.  This program recognizes that significant innovation often occurs outside large corporations.  Over ten years, the Google accelerator has supported 2,011 startups in 88 countries, achieving a 93% survival rate.  The program focuses on deep technical problem-solving with Google engineers and product managers.  It is evolving to address new areas like sovereign-level policy and frontier tech.  New initiatives are supporting green infrastructure and AI for robotics and biodiversity.  Regional alumni networks have been unified to foster a stronger global community.  Demo Days showcase startup progress, with upcoming events for various regional accelerators.  Founders can apply for support to scale technical infrastructure and optimize product-market fit.
https://cloud.google.com/blog/topics/developers-practitioners/scaling-the-next-generation-of-global-innovation-how-google-supports-top-startups-around-the-world/ cloud.google.com
RSS Hunter RSS Hunter • Jun 18

Agent Factory Recap: 100X engineering with AI agents in Google Antigravity 2.0

This discussion with Rody Davis, a top agentic engineer at Google, explores the transition to agent-first platforms like Antigravity 2.0. Antigravity 2.0 is a comprehensive platform with a desktop manager, CLI, SDK, and IDE, allowing developers custom environments. Davis emphasizes that AI accelerates the entire software lifecycle, not just coding, by reducing "toil" and allowing focus on high-level tasks. He uses "Skills," essentially cheat sheets for agents, to provide specific context like design systems or API documentation for faster, more accurate results. Customizations in Antigravity 2.0 enable extensions like Android CLI or Model Context Protocol servers for features like hot reloading. Davis compares code maintenance to Bonsai artistry, advocating for flat architectures that simplify agent guidance and human oversight. He tailors code reviews based on the task, focusing on visual output for marketing sites and API contracts for backend logic. Davis still writes code by hand to deepen his understanding of fundamental concepts. His personal website showcases offline content recommendations using Gemma 4 and vectorized summaries. A demo illustrated multi-agent parallelism where parallel sub-agents built and localized a full-stack app. The unbundling of the IDE from the Agent Manager offers workflow flexibility between desktop and server environments. Davis transforms documentation into reusable skills by parsing websites into markdown. He predicts a non-technical founder will launch a company using "vibe coding" by 2026, leading to new consulting roles addressing resulting production failures. Davis argues that poor codebase health, not context windows, is the primary bottleneck in AI development speed. He advises engineers to embrace AI for better communication handoffs, making code artifacts easily approvable. The era of agentic engineering demands increased architectural discipline, using agents as an orchestra to overcome "toil" and build future frameworks.
https://cloud.google.com/blog/topics/developers-practitioners/agent-factory-recap-100x-engineering-with-ai-agents-in-google-antigravity-20/ cloud.google.com
CdXz5zHNQW_WvoTLrvHqT.jpeg
RSS Hunter RSS Hunter • Jun 18