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

Introducing new self-learning context enrichment in Knowledge Catalog

Google Cloud's Knowledge Catalog is enhancing its context engine for agents with new preview features. Self-learning context enrichment analyzes agent and user interactions to automatically identify and update missing or outdated business context. Knowledge Studio provides a visual context graph and conversational exploration for managing this context. This addresses the challenge of manually curating context as the number of agents scales. Gemini-powered enrichment extracts verified definitions from various business documents and code repositories. These definitions are then organized into governed, reusable context sets for agents. This improves agent accuracy, reduces latency, and token usage significantly. Enrichment capabilities connect to both structured and unstructured data sources. Proposed context updates are presented with source citations and confidence scores for review. User and agent interactions pinpoint areas where agents struggle, enabling automatic fixes. Automation can be configured with confidence score thresholds for auto-publishing changes. Context sets allow for curated, domain-specific subsets of the broader context graph. These sets enforce data governance and fine-grained access control. Knowledge Studio transforms Knowledge Catalog into an interactive command center for context management. It offers visual navigation, conversational exploration, and version control for changes. These new features aim to keep context current and improve agent performance across organizations.
https://cloud.google.com/blog/products/data-analytics/knowledge-catalog-self-learning-context-enrichment-for-ai-agents/ cloud.google.com
CdXz5zHNQW_xA5im0bBF2.gif
RSS Hunter RSS Hunter • Today

Inside Agent Compute Engine: Building the compute foundation for the agentic era

The software industry faces periodic architectural shifts, with warehouse-scale computing and containers emerging after the web scale boom, and TPUs developed due to deep learning demands. Google also introduced Titanium for hardware-rooted security guarantees. A new fundamental shift is the rise of agentic AI, where AI agents take action beyond just answering questions. These agents are increasingly writing code, running tools, and orchestrating workflows, rapidly consuming compute and storage resources. Traditional compute primitives are insufficient for running millions of these agents cost-effectively and with low latency. This necessitates a new form factor, leading to the development of Agent Compute Engine, a Google Cloud service providing hardware-isolated microVMs for AI agents.Agent Compute Engine aims to bridge the gap between the isolation of virtual machines and the speed and density of containers. Existing VMs are too slow to start and stop for frequently pausing and resuming agents. Containers, while fast, lack the required isolation and kernel-level control for certain agent tasks. The system needs to efficiently manage agents, understanding when they are active or idle to avoid resource waste and preserve working state. Agent Compute Engine is designed as a unified platform offering a microVM abstraction for persistent, hardware-isolated computing.This new architecture supports diverse workloads, including interactive agent interfaces, autonomous software engineering, large-scale evaluations, and lightweight sandboxes for Google services. Security is paramount, with a defense-in-depth approach anchored in hardware-rooted trust via Titan silicon and Titanium IPUs. Host integrity is maintained through stateless, immutable appliances, and agentic sessions are isolated within single-tenant microVMs. A zero-secrets architecture prevents sensitive credentials from residing within the sandbox, ensuring security even if the guest OS is compromised.Networking is restricted, with all traffic flowing through a mediated proxy that enforces policy-driven outbound brokering. End-to-end encryption is implemented for both data in transit and at rest. Automated patching, runtime detection, and continuous security measures are in place, alongside strict separation between the agent harness and sandbox. The system operates under the assumption that the sandbox is controlled by an attacker, thus holding no credentials and having no direct network access. Governance is handled through first-class agent identities, managed via a registry and an authentication manager that acts as a vault and broker for agents.
https://cloud.google.com/blog/topics/ai-infrastructure/agent-compute-engine-deep-dive/ cloud.google.com
CdXz5zHNQW_XUm165nYZ0.png
RSS Hunter RSS Hunter • Today

Introducing Agent Compute Engine and expanded agent governance

Scaling autonomous agents like Gemini requires robust infrastructure for execution and control. Google is launching Agent Compute Engine, a specialized compute service designed for the dynamic needs of AI agents, offering isolated runtime security and efficiency. This service powers Gemini for work tasks and will soon be available to customers in preview. An accompanying open-source option, Agent Substrate on GKE, provides customization and multi-cloud portability for those standardizing on Kubernetes. Both solutions offer enhanced security through hardware isolation and rapid start-up/pause capabilities. Agent Compute Engine utilizes microVMs with copy-on-write memory sharing and fast checkpointing for efficiency, while Agent Substrate leverages a zero-trust architecture for high-density agent deployment. To address the challenges of scaling agent fleets, Google is extending Gemini for business's agent governance capabilities. Agent Gateway acts as a central policy enforcer, Agent Registry manages approved agents, and Agent Identity provides secure identification. These governance features will now cover a broader range of Google Cloud runtimes, cross-cloud environments, and even local endpoints. New governance enhancements include granular cost controls, intelligent model routing, and expanded security partner integrations. The company aims to provide a resilient agent infrastructure that combines high-performance execution with comprehensive governance for AI agents.
https://cloud.google.com/blog/topics/ai-infrastructure/introducing-agent-compute-engine-and-expanded-agent-governance/ cloud.google.com
RSS Hunter RSS Hunter • Today

ZenteiQ scales batch inference 4x with GKE Inference Gateway powered by llm-d

ZenteiQ, an AI startup from the Indian Institute of Science, is building sovereign scientific foundation models. These models require understanding of physics, equations, and numerical solvers, differentiating them from general language models. ZenteiQ employs a large-scale synthetic data generation pipeline to pre-train these models. The pipeline generates billions of reasoning tokens involving complex scientific derivations and solver validation.Previously, ZenteiQ faced routing bottlenecks on Kubernetes, leading to uneven engine queues and redundant computations. This significantly slowed down data generation and delayed pre-training. To overcome these issues, ZenteiQ implemented Google Kubernetes Engine (GKE) Inference Gateway, powered by llm-d. This solution provides intelligent, load- and cache-aware routing for their inference requests.The GKE Inference Gateway monitors real-time queue depths and KV cache utilization across their serving nodes. It routes requests effectively, preventing bottlenecks and reusing cached prompt prefixes. This intelligent routing dramatically increased throughput, with daily output growing from 1.47 billion to nearly 6 billion generated tokens. It also ensured predictability in their pre-training runs. ZenteiQ plans to leverage this setup for future advanced AI training stages.
https://cloud.google.com/blog/products/containers-kubernetes/zenteiq-scales-physical-ai-inference-4x-with-gke-and-llm-d/ cloud.google.com
CdXz5zHNQW_6aNuYU90ce.png
RSS Hunter RSS Hunter • Yesterday

BigQuery ObjectRef now GA: Build multimodal data platforms for conversational analytics entirely in SQL

BigQuery now offers ObjectRef and multimodal data analytics, turning it into a comprehensive agentic data platform. ObjectRef allows unstructured data from Cloud Storage to be represented as native columns within BigQuery tables. This integration enables direct access to unstructured files alongside structured data for analysis within a single SQL or natural language query. Users can leverage BigQuery's serverless compute and built-in inference capabilities to build complex data pipelines without separate infrastructure. Extracted features and annotations can be stored with the original unstructured data, preserving nuance for deeper analysis. This approach eliminates data-stitching overhead and database misalignment previously encountered. ObjectRef supports independent lifecycles, zero-copy architecture, and unified governance for both structured and unstructured data. Proven use cases include content moderation, threat detection, inventory compliance, and customer experience analysis. New GA features include flexible authentication, unified metadata support, and array support for comparative analytics. Companies can now consolidate their infrastructure and bring AI directly to their storage for robust multimodal analytics.
https://cloud.google.com/blog/products/data-analytics/analyze-unstructured-data-in-bigquery-with-objectref/ cloud.google.com
CdXz5zHNQW_rd8H1uSuzS.png
RSS Hunter RSS Hunter • Yesterday

Responding to core PHP vulnerabilities CVE-2026-109618 and CVE-2026-109619

Two critical PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619, have been disclosed, posing risks of unauthenticated remote code execution and memory disclosure. These flaws, with a CVSS score of 9.2, affect PHP versions 5.4 through 8.5.11 according to Wiz Research and PHP. The vulnerabilities stem from issues with out-of-memory handling and stale internal request state. Google Cloud is advising its users to immediately update their PHP applications to version 8.2 or newer and redeploy them. Specific guidance is provided for App Engine, Cloud Run, and Google Kubernetes Engine users. While Google Cloud itself is not vulnerable, customer applications running PHP on its platform may be affected. The PHP project has released patches for newer versions, but older, end-of-life versions will not be fixed. Google Cloud is rolling out automatic patches for certain serverless platforms. Users with custom container images must rebuild and redeploy them with patched PHP versions. Monitoring tools like On-Demand Scanning AP and Security Command Center can help identify and detect vulnerable workloads. Updating PHP is crucial to secure applications and maintain service integrity.
https://cloud.google.com/blog/products/identity-security/responding-to-core-php-vulnerabilities-cve-2026-109618-cve-2026-109619/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

Google Public Sector increases partner investments for frontline adoption

Google Public Sector is enhancing its partner initiatives to drive collaboration and expand mission delivery for government and education clients. These updates focus on four key areas to support their valuable partner network. New incentives will reward partners for pre-sales technical expertise and solution design. A partner-led user success program will boost Gemini adoption through targeted investments.Google is introducing a Forward Deployed Engineer academy to train elite engineers within partner organizations. Pre-sales acceleration efforts include increased funding for partner-led customer workshops and proofs of concept. Post-sales adoption sees a significant investment of $45 million in funding for partner professional services to deploy Gemini across the U.S. public sector.To support specialized missions, a new ISV Agent Factory will speed up the creation of tailored public sector agents. This complements the ISV ATO Accelerator, offering substantial funding for accreditations. Strategic alliances are being formed, including a new one with SHI to engage their sales teams and a collaboration with Accenture Federal Services for Gemini Launchpads. These initiatives aim to move agencies from experimentation to mission execution. Ultimately, these efforts ensure public sector customers have the mission-ready capabilities they need.
https://cloud.google.com/blog/topics/public-sector/google-public-sector-increases-partner-investments-for-frontline-adoption/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

Google Cloud expands model choice and flexibility, bringing OpenAI models to customers

Google Cloud is enhancing its AI offerings by integrating OpenAI models into its Model Garden. This move makes Google Cloud the sole hyperscale provider with access to all three major closed-source LLMs, alongside numerous other models. The platform emphasizes providing customers with flexibility to choose the right model for specific workloads, balancing cost, capability, and performance. New OpenAI models like GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are being introduced, each tailored for different use cases and efficiency levels. This integration allows customers to dynamically swap models without architectural lock-in or separate vendor contracts. User activity will be consolidated into a single Google Cloud invoice, utilizing existing cloud commitments and Flexible Savings Plans. Security and data governance are paramount, with strict data perimeters ensuring customer data is never used for model training. Isolated access boundaries and unified auditing provide robust security and visibility for enterprise workloads. These new OpenAI models are currently in private preview for select enterprise customers. Future integration with Gemini for business and Google Antigravity will further expand choice and efficiency for developers and enterprises.
https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-brings-openai-models-to-customers/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

Hardware-backed security at global scale: Confidential G4 with NVIDIA Blackwell is GA

Google Cloud has launched Confidential G4 instances, a significant advancement in securing artificial intelligence workloads. These instances, powered by NVIDIA RTX PRO 6000 Blackwell Server edition GPUs, offer confidential computing capabilities. This means sensitive data and intellectual property processed by autonomous agents are protected in real time. The technology integrates NVIDIA Confidential Computing into Google Cloud's AI Hypercomputer, ensuring privacy during computation without sacrificing performance. Confidential G4 instances are now generally available across numerous zones, supporting various consumption models on Compute Engine VMs and GKE nodes. Support for Confidential Space, enabling secure multi-party data sharing, is also rolling out soon. Confidential G4 instances utilize hardware-based security features like non-exportable keys, remote attestation, and memory encryption. This allows organizations to meet strict regulatory requirements and maintain cost-efficiency. The solution provides end-to-end confidentiality, isolating data and model parameters from host administrators and cloud providers. Verifiable cryptographic trust is established through hardware-rooted remote attestation, ensuring integrity before sensitive operations commence. Crucially, no code refactoring is necessary for deployment, simplifying the adoption process. Performance testing indicates that Confidential G4 instances offer throughput comparable to standard G4 instances, with minimal performance degradation. This advancement empowers enterprises to scale AI deployments securely, maintaining control, confidentiality, and accountability over their valuable AI assets.
https://cloud.google.com/blog/topics/ai-infrastructure/confidential-g4-gpus-are-ga/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

Google Cloud and NVIDIA expand co-engineering collaboration to scale infrastructure for the agentic era

Enterprises are transitioning to complex agentic AI workflows, demanding infrastructure that offers both raw performance and essential flexibility. This new era requires the ability to run diverse models for specific tasks and maintain data and operational sovereignty. Google Cloud, in collaboration with NVIDIA, is introducing advancements to meet these challenges. New A5X instances, powered by NVIDIA's Vera Rubin platform, provide accelerated computing and expanded global availability. Virgo Network enhances scale-out capabilities, enabling massive GPU clusters across sites. Storage solutions now include Managed Lustre with NVIDIA GPUDirect Storage for direct GPU data access. Advanced KV caching on A5X instances improves long-context memory for agentic workloads. Google Kubernetes Engine (GKE) will orchestrate these A5X instances, simplifying distributed AI operations. Support for open models like Gemma and Mistral is expanding within Google Cloud's Sovereign Cloud portfolio. NVIDIA's Nemotron models and Cosmos simulations are integrated with Gemini for customization and virtual world generation. Sovereign AI capabilities are being extended to on-premises environments via Google Distributed Cloud. This includes NVIDIA Blackwell Ultra GPUs and Gemini Flash models for air-gapped deployments. Confidential Computing in the public cloud is enhanced with Confidential G4 instances and Confidential Bare Metal options. These innovations aim to protect sensitive intellectual property and user data through hardware-enforced security. The co-engineered Google Cloud AI Hypercomputer with NVIDIA's full-stack platform offers a unified foundation for scaling AI.
https://cloud.google.com/blog/topics/ai-infrastructure/google-cloud-infrastructure-at-nvidia-gtc-berlin-2026/ cloud.google.com
RSS Hunter RSS Hunter • Yesterday

Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity

Google Cloud now offers Claude Opus 5.5 and Claude Sonnet 5.5 through Antigravity, providing developers with more model choices. Claude Opus 5.5 excels at complex tasks, while Sonnet 5.5 offers speed and cost-effectiveness. Administrators can enable these models alongside Google's Gemini models within Antigravity. Developers can then select their preferred models directly from the model picker across various Antigravity tools. This integration enables a unified billing and security environment for model selection. The platform supports multi-model engineering patterns to enhance code quality and manage costs. One pattern involves using different models for implementation and validation to prevent model-specific biases. Developers can assign Claude Opus 5.5 or Gemini 4 Argon for implementation and a different model for validation. Another pattern pairs powerful reasoning models with faster "Flash" subagents for efficiency. Subagents can be configured to use specific model tiers, like Gemini Flash, for high-volume tasks. This allows for optimized workflows by delegating tasks to specialized models. Administrators can configure billing and authorize partner models for their organizations. Developers should update their Antigravity tools to access these new model capabilities.
https://cloud.google.com/blog/products/ai-machine-learning/claude-opus-55-and-claude-sonnet-55-available-in-antigravity/ cloud.google.com
RSS Hunter RSS Hunter • Oct 20

TorchTPU is now open source: Run native PyTorch on TPUs

PyTorch Conference 2026 introduced new products and integrations for running PyTorch on Google Cloud and other infrastructures. TorchTPU, the native PyTorch stack for Google TPUs, is now open source on GitHub. This allows seamless training, serving, and deployment of models on TPUs using familiar tools like vLLM, SGLang, and Hugging Face. The integration provides users with a choice of computing options including TPUs, GPUs, and CPUs with a consistent operational model. TorchTPU enables running existing PyTorch workloads on TPUs with minimal code changes and familiar APIs. It supports eager mode debugging and optimizes TPU utilization and memory through torch.compile. Hugging Face models now run natively on TorchTPU, simplifying deployment with device_map="tpu". Inference optimization is achieved through TorchTPU integrations with serving engines and fleet routing across TPUs and GPUs. vLLM provides high-throughput serving and memory efficiency on TPUs with features like continuous batching and quantization. SGLang accelerates agentic and structured workloads on TPUs by reusing KV cache across requests. llm-d 1.0, a CNCF Sandbox project, connects PyTorch serving engines with Kubernetes for distributed LLM inference. TorchTitan offers PyTorch-native distributed high-performance training at scale, validated on thousands of TPUs. VeRL, a popular RL framework, now includes a TPU backend for scalable post-training and reinforcement learning workloads. Ray has added native TorchTPU support, enabling orchestration of AI pipelines across various accelerators within a single workflow. LiteRT, Google's on-device AI inference solution, now joins the PyTorch ecosystem, allowing PyTorch models to run on diverse devices. The TPU agent suite is also in early preview on GitHub to automate onboarding, diagnostics, and performance tuning for PyTorch models on TPUs.
https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus/ cloud.google.com
CdXz5zHNQW_YCinlopqHk.jpeg
RSS Hunter RSS Hunter • Oct 20

Announcing llm-d 1.0: Open distributed inference for agents on Google Cloud TPUs and GPUs

Today marks the release of llm-d 1.0, an open-source, Kubernetes-native stack for distributed LLM inference, developed by a community including Google, Red Hat, and IBM. This project integrates the PyTorch and Kubernetes communities to address evolving inference needs. Modern agent-based workloads, unlike simpler chat interactions, involve multi-step processes, tool usage, and bursty traffic patterns. Serving these on single machines or with basic load balancers proves inefficient, leading to duplicated computations and stalled interactive users. llm-d transforms clusters into coordinated serving pools, demonstrated with 1,000 agents on a single endpoint. Key features include cache-aware routing and tiered KV cache to maintain agent context and flow control to prioritize interactive requests during traffic bursts. Version 1.0 offers multi-tenant quality of service through priority-aware flow control and a stable foundation with hardened APIs. It also enables safer upgrades with high availability and zero-downtime updates, alongside agent-aware serving capabilities for context reuse. Expanded TPU integration provides native PyTorch model serving with TorchTPU time-slicing for improved throughput and utilization. llm-d components are composable, allowing users to adopt only necessary serving elements. This stack connects Kubernetes primitives, PyTorch serving engines, and accelerator-aware scheduling for a flexible production environment.
https://cloud.google.com/blog/topics/ai-infrastructure/llm-d-1-0-an-open-source-stack-for-distributed-inference/ cloud.google.com
CdXz5zHNQW_U6mQXXQiGC.png
RSS Hunter RSS Hunter • Oct 20

Announcements and highlights from Google Public Sector Summit 2026

Google is hosting a summit to discuss how public sector missions can achieve ambitious goals through technology. Public sector leaders face increasing demands for speed and efficiency amidst complexity and limited resources. While many agencies have adopted AI agents, widespread adoption is hindered by legacy data silos and integration challenges. Google aims to simplify these processes with secure, scalable technology.The company is introducing the Gemini agent, a universal assistant for various work tasks that integrates with existing systems and adheres to organizational security controls. Gemini for government services is a new platform designed for public sector security and workflow requirements, enabling secure connections to existing systems without data migration. Wiz agents are now available within Wiz for Gov, a FedRAMP certified environment, to accelerate remediation for critical priorities like Post-Quantum Cryptography. Wiz also offers no-cost security assessments through its Scan for Good program.Enterprise search capabilities are being brought to Gemini on Google Distributed Cloud for air-gapped environments, allowing classified data analysis without internet compromise. Partnerships with organizations like America.gov, SUNY, the Department of Energy, the U.S. Army, and the Department of War are demonstrating how Google's technology drives mission impact. These collaborations focus on simplifying federal service access, advancing research, strengthening national security, and improving logistics. Case Western Reserve University is using Gemini to enhance learning and research. Google emphasizes its role as a mission partner in the agentic era, encouraging public sector entities to build and create impact together.
https://cloud.google.com/blog/topics/public-sector/announcements-and-highlights-from-google-public-sector-summit-2026/ cloud.google.com
RSS Hunter RSS Hunter • Oct 20

Introducing Gemini for government services

Google has launched Gemini for government services, an AI agent designed for public sector organizations. This universal agent handles knowledge work, answers questions, creates media, and runs code based on stated objectives. It connects to existing agency systems without data migration and adheres to strict organizational security and compliance requirements. The government version includes specialized skills for public service, like policy research and FOIA discovery, and integrates with systems such as Oracle and Microsoft 365. Pre-built partner agents and system integrators are available for customization and deployment. Gemini for government services operates on Google Cloud, ensuring data privacy and U.S. data residency. It aims to automate routine tasks, freeing up civil servants for more critical work. The preview version is part of Google's broader strategy to tailor Gemini for specific industries. This initiative seeks to accelerate complex government operations and improve citizen services.
https://cloud.google.com/blog/products/ai-machine-learning/gemini-for-government-services/ cloud.google.com
RSS Hunter RSS Hunter • Oct 20

Managed RAG search inside the sovereign perimeter: Gemini on GDC enterprise search

Organizations handling sensitive data like classified reports require secure data analysis solutions. Public cloud search capabilities are often restricted due to data sovereignty mandates preventing data from leaving premises. Previously, achieving this involved complex, time-consuming custom infrastructure development for disconnected environments. Now, Google Distributed Cloud (GDC) air-gapped offers enterprise search with Gemini to address these challenges. This fully managed retrieval-augmented generation search runs entirely within a secure physical perimeter, disconnected from the internet. Users can upload files and ask Gemini to find, summarize, compare, or translate information, with cited sources and persistent conversations. This enables cross-silo data searching and unlocks insights from various formats, including audio, video, and images, without separate transcription steps. The system integrates with existing identity providers for need-to-know access, ensuring privacy. Data is encrypted at rest and in transit, with security controls aligned with government standards. Gemini's local code execution engine securely processes analytical queries from structured data. This solution brings advanced multimodal intelligence to sovereign data centers without compromising security.
https://cloud.google.com/blog/topics/hybrid-cloud/managed-rag-search-inside-the-sovereign-perimeter-gemini-on-gdc-enterprise-search/ cloud.google.com
CdXz5zHNQW_YS197tO5bU.png
RSS Hunter RSS Hunter • Oct 20

Enabling the next era of sovereign AI: The latest Google Distributed Cloud innovations

Agentic systems are increasingly valuable for enterprises, but organizations with strict data sovereignty mandates face challenges deploying them. Sending sensitive data to public clouds is often not an option, and building private AI stacks is time-consuming. Google Distributed Cloud (GDC) air-gapped innovations now address this by enabling deployment directly where sensitive data resides. Key advancements include enterprise search powered by Gemini, support for open-weight models, turnkey ISV applications, and accelerated infrastructure. GDC aims to transform isolated infrastructure into a modern agentic innovation platform with flexibility from foundational and open models. Gemini on GDC enterprise search offers a conversational interface for querying vast data repositories within sovereign perimeters. It supports multimodal data like text, audio, and video with low-latency inference and built-in retrieval-augmented generation. GDC air-gapped now generally available supports Gemini 3.5 Flash and leading open models like Gemma and Mistral, offering workload choice. The platform features a new, elevated Infrastructure-as-a-Service stack for running AI and traditional workloads together. This includes A4 Ultra machine types with NVIDIA Blackwell GPUs for high-performance agentic infrastructure and resilient, managed networking solutions. An expanding ISV partner ecosystem brings familiar software like Teradata, Solo.io, and Mattermost to run natively on GDC. These integrations allow organizations to leverage AI without compromising data privacy or leaving familiar tools behind, transforming sensitive data into actionable insights within compliant, secure, and controlled air-gapped environments.
https://cloud.google.com/blog/topics/hybrid-cloud/new-innovations-in-google-distributed-cloud/ cloud.google.com
CdXz5zHNQW_WoYmLn2p8a.png
RSS Hunter RSS Hunter • Oct 20

How to Use Python UDF Logging for Pipeline Observability

BigQuery now supports Python UDF logging, allowing developers to emit structured logs directly into Cloud Logging using Python's standard logging library. This new feature facilitates operational monitoring and debugging without external plugins, particularly useful for building fault-tolerant data pipelines. The workflow involves inline validation within the Python UDF, real-time alerting via Cloud Monitoring on log-based metrics, and SQL triage/replay of quarantined data.A practical example demonstrates an e-commerce event ingestion pipeline. The generic Python UDF parses JSON, logs diagnostic information for invalid records, and returns NULL to prevent query failure. Logging is enabled via SET @@udf.properties in the BigQuery session.Real-time alerts can be configured in Cloud Logging using a specific log filter for UDF errors. For forensic analysis, BigQuery Log Analytics over the linked _AllLogs view allows data engineers to query quarantined payloads directly with SQL. Because raw payloads are preserved, problematic records can be recovered and re-ingested into the production table, ensuring no data loss.Beyond data validation, this logging feature supports use cases like shadow execution for safe migrations, PII redaction auditing, and detecting ML inference drift. Best practices include enabling debug output, avoiding sensitive data in logs, and managing logging costs by selective logging. This integration streamlines debugging and enhances data pipeline resilience within BigQuery.
https://cloud.google.com/blog/products/data-analytics/python-udf-logging-for-pipeline-observability-in-bigquery/ cloud.google.com
CdXz5zHNQW_vlOyaZLNoM.jpeg
RSS Hunter RSS Hunter • Oct 19

Developer’s guide to implementing MCP apps in Gemini

Enterprise assistants primarily use text, which is inefficient for visual, multi-step tasks requiring frequent adjustments. MCP Apps, an extension of the Model Context Protocol, solves this by pairing tools with live user interfaces rendered directly within the Gemini Enterprise chat flow. These interactive UIs, embedded in sandboxed iframes, can expand to full screen or a side panel. Edits are processed locally, maintaining synchronization between the workspace and the assistant. This approach eliminates the need to rebuild core infrastructure, leveraging Gemini Enterprise for authentication, data access, and the default assistant. Users host their MCP server on a private Cloud Run service. Throughout generations, enterprise assistants evolved from static text chatbots to those invoking live APIs, and now to interactive MCP Apps with live workspaces. MCP Apps offer a responsive user experience, purpose-built workflows, enterprise-grade security, human-in-the-loop governance, and broad connectivity. The architecture involves three decoupled tiers: the client browser, Gemini Enterprise, and the user's Google Cloud project. Security is enforced through two layers of authentication: Cloud Run IAM and in-app OAuth. Building an MCP App involves creating a server and HTML widget, deploying to Cloud Run, and registering it within Gemini Enterprise. Once registered, MCP tools are imported as actions that can be enabled and connected to Gemini Enterprise apps for seamless integration.
https://cloud.google.com/blog/topics/developers-practitioners/devs-guide-to-mcp-apps-in-gemini/ cloud.google.com
CdXz5zHNQW_eem4fhIiKn.png
RSS Hunter RSS Hunter • Oct 19

Who judges the judge? Three helpful ways to audit your LLM evaluators

A common tech misconception is that AI offers objective evaluations, but LLMs, like humans, are subjective. LLM judges exhibit several biases, including verbosity bias, position bias, and the halo effect, where clear formatting can mislead them. They also show agreeableness bias, failing to detect errors, and self-enhancement bias by favoring their own outputs. Therefore, LLM judges are not objective and treating them as infallible risks compromised work. To improve LLM judge quality, a three-pillar method is suggested. The first pillar is measuring internal stability by checking consistency across multiple passes for identical inputs. The second pillar involves validating human alignment using chance-corrected metrics to ensure they identify real errors, not just agree with lenient scores. The third pillar advocates for using multi-judge ensembles, employing decomposed prompting for specialized scoring and running multiple models simultaneously to track performance and identify divergences. Partnering with Delta can help implement statistically-audited evaluation pipelines.
https://cloud.google.com/blog/topics/consulting/three-ways-to-audit-your-llm-evaluators/ cloud.google.com
RSS Hunter RSS Hunter • Oct 19

Building full virtual vehicles on Google Cloud: From AAOS to the smallest MCU

The automotive industry is rapidly transitioning to software-defined vehicles, moving core functions from physical hardware to digital systems. While infotainment and high-performance computing systems are often developed using cloud-native methods, many crucial microchips and control modules remain hardwired. This creates a significant bottleneck, making updates and expansion difficult post-manufacturing and pushing costly defect resolution to the end of development cycles.Google, through partnerships like with RemotiveLabs and its Horizon platform, aims to address these inefficiencies. Android Automotive OS (AAOS) already leverages virtual testing environments like Cuttlefish on Arm-based Google Cloud instances for seamless cockpit software development. RemotiveTopology extends this virtual model across the entire vehicle's electrical and electronic (E/E) architecture.By converting vehicle networks into "vehicle-as-code," RemotiveTopology creates a full, simulated E/E architecture that connects AAOS to other vehicle systems. This allows developers to spin up a complete virtual vehicle on demand, enabling continuous signal and cross-domain visibility. The system, built on standard Linux tooling, runs on various platforms, from laptops to Google Cloud.This approach facilitates early, rather than late, integration of all onboard systems within a virtual environment before physical hardware is involved. Critical, multi-domain interactions, such as ADAS events appearing in the infotainment cluster, can be tested continuously from the start. This early integration significantly reduces the likelihood and cost of discovering defects later in the development process.The ultimate goal is to enable scalable software development for even the smallest vehicle components, bridging the gap to fully software-defined vehicles. Google Cloud's Arm-based infrastructure, including Axion's bare-metal C4A instances, ensures parity between cloud-developed software and actual vehicle control modules. This holistic system allows for the first time, full vehicle system development and integration testing in the cloud, directly transferable to physical vehicles.The benefits are substantial, including over 10x faster integration cycles, as measured in early customer engagements. Virtual systems are crucial for keeping pace with AI-assisted workflows, which generate, test, and refine vehicle software rapidly. RemotiveTopology makes this architecture accessible, offering an open-source simplified vehicle on GitHub for immediate use without new third-party tooling.Manufacturers can leverage existing toolchains and easily integrate RemotiveTopology into Arm-based Google Cloud environments. For those currently limiting virtual development to the cockpit, it's time to expand to the entire vehicle in the cloud through this innovative approach.
https://cloud.google.com/blog/topics/manufacturing/building-full-virtual-vehicles-on-google-cloud-from-aaos-to-the-smallest-mcu/ cloud.google.com
RSS Hunter RSS Hunter • Oct 19

Google named a Leader in the 2026 IDC MarketScape for Agentic AI Platforms

Organizations face significant challenges in deploying AI agents across their businesses, prioritizing quality, integration, security, and governance over raw model intelligence. Gemini by Google is presented as a foundational platform designed to address these deployment hurdles. It enables agents to operate within familiar user environments, like Google Workspace and Microsoft 365, maintaining consistent memory and permissions. Gemini also champions radical openness, offering a choice of over 200 models and direct integration with enterprise data sources such as Salesforce and SAP. The platform employs smart token economics, optimizing costs by selecting the appropriate model for each task and allowing for reusable skills. This approach enhances efficiency and reduces redundant processing. Furthermore, Gemini incorporates built-in governance and security features, including dedicated agent identities, isolated sandboxes, and automatic policy enforcement. These safeguards protect sensitive data and infrastructure. The platform aims to provide a reliable and open environment for builders, fostering confidence in AI deployments. Google's leadership in the inaugural IDC MarketScape for Agentic AI Platforms underscores Gemini's capabilities in meeting enterprise AI needs. Gemini's design focuses on empowering teams with choice and robust security for AI integration.
https://cloud.google.com/blog/products/ai-machine-learning/google-named-leader-in-2026-idc-marketscape-for-agentic-ai-platforms/ cloud.google.com
CdXz5zHNQW_tDy32XzBIv.png
RSS Hunter RSS Hunter • Oct 19

Day 0 support for Reflection’s Beam on Google Cloud: 5x faster perf with llm-d and NVIDIA Dynamo

Reflection has launched Beam, an open-weight model designed for advanced software engineering and long-horizon agentic tasks. Beam operates under an Apache 2.0 license and offers impressive inference efficiency for models in its parameter class. Google Cloud announced immediate support for Beam, providing deployment recipes on their AI Hypercomputer and NVIDIA AI Infrastructure. Together with Reflection, Google Cloud optimized Beam's performance using llm-d and NVIDIA Dynamo for enhanced serving. Deploying Beam on Google Cloud resulted in significantly higher performance and lower latency for agentic workloads. The release emphasizes compounding model and system efficiency, leading to greater throughput and reduced latency for agents. High-efficiency multi-turn serving is achieved through advanced techniques like KV-cache-aware routing and speculative decoding. Developers can now benefit from turnkey optimized deployments for agentic workloads with ease. Scaling inference for agentic tasks requires optimizations at both the model and serving system layers. Beam's unique architecture, activating only a fraction of its parameters per token, drastically reduces compute requirements. The serving stack addresses bottlenecks like prefill compute and KV cache capacity in long-context scenarios. This jointly developed solution aims to empower developers with powerful and efficient AI tools.
https://cloud.google.com/blog/topics/ai-infrastructure/day-0-support-with-5x-perf-for-reflection-beam-on-google-cloud/ cloud.google.com
CdXz5zHNQW_Vmt2Gjgcaf.png
RSS Hunter RSS Hunter • Oct 16

Architecting autonomy: A design pattern for high-compute, governed agents in biotech

Modern AI agents function as simple dispatchers, quickly processing prompts and API calls. However, biotechnology tasks like protein folding and compound screening are complex, sustained simulations, unlike fast API lookups. Standard AI paradigms fail with these demanding workflows due to timeouts and IP exposure risks. Building enterprise-grade R&D agents requires a new approach, moving beyond monolithic assistants. This involves a three-layer architecture separating cognitive orchestration, computational execution, and enterprise governance.The execution layer handles heavy scientific computations, with workloads run on dedicated GPUs or managed cloud endpoints. To mitigate GPU cold-start penalties, maintaining warm pools and preloading weights are crucial for responsiveness. Containerized NVIDIA Inference Microservices (NIMs) ensure rapid prototyping while maintaining strict data isolation for proprietary chemical information. The cognitive layer orchestrates agentic workflows using a hierarchical multi-agent structure.Key orchestration patterns include in-process native agent handoffs for speed and passing bulky payloads by reference via cloud storage to keep context windows compact. Workflow logic is decoupled from biochemical reasoning, assigning planning to general LLMs and domain expertise to specialized models. Evaluation incorporates context-driven assessment rather than rigid cutoffs, allowing for dynamic risk tolerance in drug candidate assessment.Performance best practices emphasize optimizing I/O and payload sizes over raw token speed. Aligning client-side concurrency with backend mechanics and avoiding deep tool chains are also vital. Passing heavy payloads by reference and streaming intermediate stage completions improve visibility and efficiency. Wrapping end-to-end pipelines behind a single, clean interface simplifies compute.The Gemini platform provides enterprise governance with agent identity, an agent gateway for secure ingress/egress, and an agent registry for approved tools. Model Armor inspects prompts and payloads for security. Asymmetric autonomy balances reversible automated operations with gated irreversible ones requiring human sign-off. This decoupled architecture ensures scalable, secure, and auditable research systems for future complex simulations.
https://cloud.google.com/blog/topics/healthcare-life-sciences/architecting-scalable-agentic-biotech-workloads/ cloud.google.com
RSS Hunter RSS Hunter • Oct 16

Announcing Wiz Runtime Sensor add-on for GKE: Turnkey threat detection for Kubernetes

The age of AI accelerates security threats, making traditional scans insufficient for active attacks. Wiz is introducing a Runtime Sensor add-on for Google Kubernetes Engine (GKE) in preview. This integration offers low-touch deployment and automated management for Wiz's eBPF runtime threat detection. It addresses common deployment challenges like complex installations and credential security risks. The Wiz Runtime Sensor for GKE is now a first-party, managed add-on simplifying sensor lifecycle management. It seamlessly integrates with GKE's Secret Manager for secure credential handling. Platform operators can enable the sensor using gcloud CLI or the GKE API for simplified deployment. Fleet-wide governance is enforced through organization policy constraints, preventing clusters without the sensor. The sensor provides kernel-level observability with minimal overhead, monitoring system calls and network activity. It also auto-updates independently, ensuring continuous protection against emerging threats. This integration with the Wiz AI Application Protection Platform offers real-time risk assessment and threat detection.
https://cloud.google.com/blog/products/identity-security/wiz-sensor-for-gke-turnkey-threat-detection-for-kubernetes/ cloud.google.com
CdXz5zHNQW_WGMhfwSYQ0.jpeg
RSS Hunter RSS Hunter • Oct 16

The SAVE method: Learn the fundamentals to engineer trustworthy generative media

Google Cloud Consulting offers a four-step method called SAVE to create trustworthy and brand-aligned generative media. This approach moves beyond complex prompting to ensure consistent results. The first step, Separate, involves dividing the prompt into distinct roles for a researcher and a storyteller. The researcher extracts verified facts from source documents, while the storyteller focuses on creative expression using that curated information. Approving verified facts only, the second step, means the generative model should never access raw, unedited files. Instead, it should receive only approved statements and contextual packaging with explicit constraints. Verifying before rendering, the third step, emphasizes reviewing scripts and storyboards before media generation. This allows for easy edits of text-based content to catch errors early, saving time and resources. The final step, Engineer, involves building programmatic guardrails and independent safety gates into the system. This enforces rules like word counts and brand compliance deterministically, rather than relying solely on prompt obedience. By implementing SAVE, organizations can reduce the risk of fabricated details and ensure accuracy in their generative media. This method places the responsibility for accuracy within the system's architecture. Applying SAVE helps architect trust across generative media using Google Cloud tools.
https://cloud.google.com/blog/topics/consulting/fundamentals-to-engineer-trustworthy-gen-media/ cloud.google.com
RSS Hunter RSS Hunter • Oct 16

Four evaluation engineering lessons learned from building agent plugins

The real challenge in building agentic systems is not construction but measurement. Evaluation engineering is crucial for creating trustworthy AI systems by implementing systematic, repeatable tests. Developing the Google Cloud Developer plugin provided real-world insights into these evaluation principles.One key lesson is to grade execution steps rather than just the final summary. Evaluating the raw tool-call trajectory, verifying commands executed, validated, and scoped, is more accurate than relying on an LLM judge to score the agent's text reply. Another lesson emphasizes designing evaluations for asynchronous operations. Preserving turn structure in telemetry is vital to understand concurrent tool dispatch and avoid issues where guidance is processed after actions are taken.It is also important to test live runs and "dry runs" side-by-side. Live runs verify actual execution and safety, while plan-only runs reveal the model's reasoning without permission errors. The delta between these two can expose significant cognitive flaws. Finally, using "golden sets" as durable baselines is recommended. These sets represent ideal paths curated by experts and serve as a safety net against model and harness evolution.These evaluation suites are essential for enterprise-grade agents that handle multi-step interactions and permissions. Building and shipping a reliable agent requires more work in evaluations to ensure expected behavior. The durability of a stress-tested eval suite ultimately outlasts model churn.
https://cloud.google.com/blog/topics/developers-practitioners/four-eval-engineering-lessons/ cloud.google.com
CdXz5zHNQW_EljdokhaYM.png
RSS Hunter RSS Hunter • Oct 16

Bringing the borderless Lakehouse into the agentic era with innovations for Gemini

Enterprises are moving AI from chatbots to autonomous agents, but struggle with scattered data. This bottleneck prevents agents from becoming true systems of action without security risks. Gemini, integrated with a borderless Lakehouse, now offers a solution. This integration allows Gemini to access and reason over data across clouds, SaaS platforms, and on-premises systems without data movement. Business users can query diverse data sources using natural language. AI agents can leverage this data for grounded reasoning and code generation. New advancements in the borderless Lakehouse include cross-cloud support and Apache Iceberg compatibility. Performance is enhanced for petabyte-scale reasoning through dedicated network connections and caching. Optimizations for Iceberg metadata improve query engine efficiency. Read-write interoperability allows seamless work on shared Iceberg tables. AlloyDB lakehouse federation enables querying of database and warehouse data together. The Lakehouse runtime catalog unifies enterprise data, providing context for agents. Zero-copy data integration with SaaS applications is now expanded. Databricks Delta Lake assets can be synchronized into the Lakehouse. Real-world examples demonstrate significant performance improvements and modernization. Migrating from legacy systems to the borderless Lakehouse is now streamlined.
https://cloud.google.com/blog/products/data-analytics/borderless-lakehouse-integrates-with-gemini/ cloud.google.com
CdXz5zHNQW_l99CalEsKz.gif
RSS Hunter RSS Hunter • Oct 15

Temporal Serverless Workers on Cloud Run: Scale workers ahead of demand

The Temporal execution platform, crucial for distributed systems and AI agents, faces challenges in worker deployment and scaling. Traditional methods often involve managing Kubernetes or VMs, leading to continuous costs and scaling lags. Temporal Cloud now offers Serverless Workers on Cloud Run, addressing these issues. These workers are deployed as a "worker pool" that Temporal automatically scales, even down to zero when idle. This eliminates the need for managing clusters or autoscalers, keeping worker code unchanged.Cloud Run worker pools are designed for continuous background tasks, aligning with how Temporal workers operate. Temporal Cloud monitors task queues and adjusts the worker pool size via the Cloud Run Admin API. Each build version of a worker runs in its own pool, allowing older versions to scale to zero until needed. This technology enables autoscaled agentic workers for AI, handles long-running workflows efficiently, and supports bursty batch processing with cost savings.Previously, running Temporal workers on Cloud Run required workarounds for port listeners and instance billing. Worker pools remove these, offering no endpoint management and steady CPU at a lower cost. Scaling doesn't cause revision churn, and Temporal manages growth by monitoring task backlog latency. Scale-in is more conservative to ensure availability, with configurable graceful shutdown for workers.Temporal Serverless Workers on Cloud Run also enforce Temporal Worker Versioning, a critical feature for stateful, deterministic workflows. Deploying new code involves creating distinct worker pools for each version, preventing mid-run errors. Workflows remain on their original version or can be configured for automatic upgrades, depending on the workflow type. Setting up involves containerizing workers, deploying them as a Cloud Run worker pool, granting Temporal Cloud permissions, and creating a Worker Deployment Version. This new feature is currently in preview.
https://cloud.google.com/blog/products/serverless/temporal-serverless-workers-on-cloud-run-worker-pools/ cloud.google.com
CdXz5zHNQW_Mquaroke6P.png
RSS Hunter RSS Hunter • Oct 15

How Visko streams real-time world models on Google Cloud TPUs

Generative AI startups are now developing interactive, real-time worlds instead of just static clips. Visko collaborated with Google Cloud TPU engineers to deploy its live world model, Orbis, on Cloud TPUs in just 30 days. Serving a live world model presents a unique challenge of beating a strict real-time clock, unlike offline video generation. Every video and audio chunk must render before the previous one finishes to avoid freezing the world. Visko's Orbis streams live video from a single prompt and allows viewers to dynamically steer the scene in real time. The teams successfully deployed Visko's end-to-end inference pipeline, including video generation, upscaling, and audio, onto a single Cloud TPU slice. They achieved end-to-end generation at 1.07x playback speed through architectural optimizations and precise engineering. This technical deep dive explores the optimizations and practical lessons for startups building ultra-low-latency generative AI on Google Cloud. Live world models are fundamentally different from standard batch inference due to a batch size of one and strict sequential dependencies. Visko's pipeline chains three generative models on one Cloud TPU slice, sharing hardware for each chunk.
https://cloud.google.com/blog/products/media-entertainment/how-visko-streams-real-time-world-models-on-google-cloud-tpus/ cloud.google.com
CdXz5zHNQW_qZss1VvSMS.png
RSS Hunter RSS Hunter • Oct 15

Modernizing perimeter security: Scalable perimeters with VPC Service Controls

Autonomous agents in cloud operations present new security challenges due to their non-human identity and ability to act at scale. Traditional network perimeters are insufficient against machine-speed attacks and can be too restrictive or permissive. Enhanced VPC Service Controls (VPC-SC) now offer easier perimeter creation, maintenance, and scaling. Folder membership in VPC-SC allows projects within folders to inherit security policies automatically. Granular exceptions and guided remediation for denied access are now available. This ensures data is protected by default, enabling the scaling of autonomous agents securely. VPC-SC provides context-aware network security perimeters against data theft and insider threats. Folder and organization support in ingress/egress rules simplifies management and keeps rules concise. Private Service Connect endpoints can now be used in ingress/egress rules for more precise policy authoring. Granular control over Google API access is offered through service patterns, allowing specific URL patterns. Guided violation remediation streamlines the process of authorizing legitimate access requests, reducing resolution time significantly.
https://cloud.google.com/blog/products/identity-security/introducing-new-scalable-perimeter-options-with-vpc-service-controls/ cloud.google.com
CdXz5zHNQW_XbGjrKCqAa.png
RSS Hunter RSS Hunter • Oct 15

What the Gemini agent means for startups

Google for Startups has introduced the Gemini agent to help lean startup teams overcome operational overhead and focus on core product development. This unified, omnipresent, and persistent agent acts as a personal assistant and a dedicated team member. It handles diverse tasks like writing, coding, and data analysis, dynamically creating sub-agents for complex jobs. The Gemini agent integrates across daily tools and cloud services, maintaining context over long, multi-step processes. Founders can leverage it like a digital chief of staff, automating tasks from market research to pitch deck creation.Key benefits for startups include scaling output with coworker agents and reusable skills. They can choose from various AI models, optimizing costs without vendor lock-in. The agent connects seamlessly with existing engineering, data, and business tools, reducing the need for custom connectors. Built-in enterprise-grade security and governance features, including agent identity and isolated sandboxes, help B2B startups clear procurement hurdles. Real-time cost controls and spend caps protect runway by preventing unexpected billing surprises. Early adopters like Whenere are already using the Gemini platform to create assets efficiently. Startups can contact Google for Startups to learn more about these capabilities.
https://cloud.google.com/blog/topics/startups/what-the-gemini-agent-means-for-startups/ cloud.google.com
RSS Hunter RSS Hunter • Oct 15

ARMORY CASTLE Exploits BeyondTrust

BeyondTrust disclosed a vulnerability, CVE-2026-102764, in its Remote Support appliances, allowing arbitrary file uploads. Mandiant discovered suspected PRC-nexus actor ARMORY CASTLE exploited this as a zero-day in late August 2026. The actor established a foothold using a custom PHP webshell called SAWFLY, which communicated via encrypted C2. Privilege escalation to root was achieved by manipulating SUID permissions on a system binary. ARMORY CASTLE then harvested credentials using the proprietary bt-secrets CLI utility. For long-term persistence, they modified appliance configuration templates to grant passwordless root access to a service account. The threat actor actively deleted logs to evade detection and hinder forensic analysis. They also abused compromised administrator credentials to gain access to the BeyondTrust web portal. Furthermore, ARMORY CASTLE modified user permissions to permanently gain unrestricted remote access to downstream systems. Finally, they used a custom tunneler named BARECLAW to exfiltrate data and move laterally within the network.
https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-beyondtrust/ cloud.google.com
CdXz5zHNQW_6mJOqwr4n6.jpeg
RSS Hunter RSS Hunter • Oct 15

Investigating Ephemeral Supply Chain Tampering

Google Threat Intelligence Group and Mandiant investigated a software supply chain attack against AOMEI Technology. The threat actor, UNC7017, repeatedly tampered with AOMEI's update servers in very short windows of five to ten minutes. This tactic was designed to evade traditional security detection methods. The campaign delivered a multi-stage payload, including the CROSSROAD downloader and SHINYROAD backdoor. UNC7017 strategically shifted its infrastructure from cloud providers to dedicated domains and cloud storage. Organizations across various sectors in multiple countries were impacted by this opportunistic campaign. The ManagedEngine Threat Analysis Lab (METAL) initially detected the activity through their EDR telemetry. GTIG subsequently notified affected organizations and shared findings with AOMEI, who then took corrective actions. The attack exploited AOMEI's update and notification logic by modifying configuration files like down_html.ini and Notifycfg.ini. UNC7017 also used a silent WebDAV credential coercion technique with an invisible popup. Malware delivery involved custom loaders and post-exploitation frameworks, including a modified Radmin Server backdoor. The CROSSROAD downloader validated the environment, cleaned endpoints, staged secondary payloads, and established persistence.
https://cloud.google.com/blog/topics/threat-intelligence/investigating-ephemeral-supply-chain-tampering/ cloud.google.com
CdXz5zHNQW_qi7EFuLDav.png
RSS Hunter RSS Hunter • Oct 15

Predictable AI spend in BigQuery: Introducing cost management for generative AI functions

BigQuery generative AI functions allow users to integrate large language models directly into SQL queries for tasks like classification and sentiment analysis. These functions are billed per token, making proactive budget management crucial for large-scale use. Google Cloud has introduced token-based cost management for these functions through daily quotas managed via IAM.Four daily quotas exist: AI input tokens per day, AI input tokens per day per user, AI output tokens per day, and AI output tokens per day per user. These quotas apply globally and cover specific generative AI functions in BigQuery, excluding embedding functions. Administrators can view and adjust quotas directly in the Google Cloud console or programmatically through APIs.To manage costs effectively, users can lower quotas immediately or request increases, which are often auto-approved. Monitoring usage on the Quotas & System Limits page helps detect spikes and right-size budgets. Token quotas act as a preventive, AI-specific guardrail, distinct from broader Cloud Billing spend caps.Enforcement involves stopping work when quotas are reached, either rejecting queries upfront or halting LLM calls mid-query. The AI.COUNT_TOKENS function aids in estimating input token usage without incurring extra costs. By understanding token pricing and estimating usage, administrators can set appropriate quota values to align with budget goals.Best practices include starting with per-user quotas, measuring input data with AI.COUNT_TOKENS, being mindful of the higher cost of output tokens, and iteratively adjusting limits based on observed usage. Implementing these token-based guardrails provides cost predictability for generative AI in BigQuery.
https://cloud.google.com/blog/products/data-analytics/using-bigquerys-generative-ai-cost-management-features/ cloud.google.com
CdXz5zHNQW_TKvYyrBXNC.png
RSS Hunter RSS Hunter • Oct 14

Cloud CISO Perspectives: New Gemini Notebook can help CISOs and policymakers prepare for PQC

Google Cloud has launched a public Gemini Notebook to help CISOs and executives prepare for post-quantum cryptography (PQC). This tool consolidates Google's PQC research, engineering insights, and migration advice into an accessible format. Quantum computing is advancing rapidly, making current encryption vulnerable to future attacks like "store-now-decrypt-later." Google estimates that breaking common encryption schemes would require a significant number of quantum computer components. The company plans to transition to PQC by 2029 and urges other organizations to do the same. The Gemini Notebook offers an audio overview, flashcards, and answers to specific questions about PQC. CISOs are advised to identify vulnerable cryptography, prioritize risks, and build crypto-agility into their plans. Google is already deploying PQC in products like Chrome and Android and is a co-founder of the Post-Quantum Cryptography Alliance. Organizations can create their own PQC notebooks by curating trusted sources and using neutral prompts. The article also highlights recent security updates and threat intelligence, including vulnerability trends in the AI era and exploitation of network appliances. Podcasts from Google Cloud discuss AI in browser patching and vulnerability research.
https://cloud.google.com/blog/products/identity-security/cloud-ciso-perspectives-new-gemini-notebook-can-help-cisos-policymakers-prepare-for-pqc/ cloud.google.com
CdXz5zHNQW_EUDsV4t64q.jpeg
RSS Hunter RSS Hunter • Oct 14

Cross-cloud conversational analytics made simple with open standards

Organizations often struggle with fragmented data spread across multiple specialized platforms and clouds. Previously, this required either building complex data pipelines or accepting incomplete reporting. Traditional methods involve copying data into central warehouses, leading to synchronization issues, storage costs, and broken pipelines when schemas change. Google Cloud's borderless Lakehouse, built on Apache Iceberg, offers a solution by enabling compute and AI to access data in place across different platforms. This architecture synchronizes metadata while allowing direct access to data files. It supports federating data from Databricks, Snowflake, and AWS Glue into BigQuery without data migration. Federated queries in BigQuery use a four-part path to access data across these external catalogs. This integration allows for unified querying and analysis of disparate datasets. Furthermore, federated tables can be used with BigQuery's conversational analytics for exploration using natural language. Operational considerations include managing background refresh timing, understanding query caching, and ensuring data residency compliance. The borderless Lakehouse aims to simplify data activation by eliminating costly migrations and complex pipelines.
https://cloud.google.com/blog/products/data-analytics/cross-cloud-conversational-analytics-with-borderless-lakehouse/ cloud.google.com
CdXz5zHNQW_LuveKFuJPr.png
RSS Hunter RSS Hunter • Oct 13

Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems

Gemini Enterprise has been updated with the Gemini agent, a universal AI assistant designed for work. This agent is always available to answer questions, write code, and complete complex tasks from a single prompt. Underpinning this is Gemini 4 Argon, Google's new frontier model, now available on Google Cloud. Argon is trained for depth and precision, adept at maintaining context across long, multi-step tasks, and excels in areas like software builds, finance, legal work, and cybersecurity. It transforms unstructured inputs into cohesive outputs, handling documents with text, tables, charts, and images. Argon also delivers precise and verifiable results, automating financial reconciliations, legal research with verified citations, and improving task success rates in areas like HR and customer service. The model helps build and ship products faster by breaking down goals, tracking progress, and updating files simultaneously, leading to significant performance improvements in code development and architectural refactoring. In cybersecurity, Argon offers defensive capabilities through tools like CodeMender to find and fix vulnerabilities and exhibits resilience against prompt injection attacks. Gemini 4 Argon is accessible through Gemini for business and via API for developers. Introductory pricing is available until November 30, 2026.
https://cloud.google.com/blog/products/ai-machine-learning/gemini-4-argon-available-on-google-cloud/ cloud.google.com
CdXz5zHNQW_WbMOCLyl5W.png
RSS Hunter RSS Hunter • Oct 13

One Agent Development Kit, five languages

Modern agents can be built in various languages, not just Python, but many frameworks assume a Python-only stack. Google's Agent Development Kit (ADK) aims to eliminate this by supporting Python, Go, TypeScript, Java, and Kotlin. Each ADK version is tailored for its runtime while sharing a unified architecture and interoperability via open protocols. The Model Context Protocol (MCP) client in every language allows custom tools to be called universally. For multi-agent systems, the Agent2Agent (A2A) protocol enables communication between agents written in different languages. Getting started involves a one-time setup and scaffolding a project with deployment configurations. The ADK offers distinct advantages for each language, from Python's broad integrations and evaluation tooling to Go's production efficiency and concurrency. TypeScript leverages Zod for type safety and can run directly in the browser. Java integrates with existing Maven, Gradle, and Spring Boot ecosystems. Kotlin prioritizes build-time safety with its KSP compiler plugin, enabling on-device Android agent development. The ADK's multi-language support allows organizations to select the best language for each component without compromise.
https://cloud.google.com/blog/topics/developers-practitioners/one-agent-development-kit-five-languages/ cloud.google.com
CdXz5zHNQW_9VOuIRp4Dk.jpeg
RSS Hunter RSS Hunter • Oct 13

10 Years of the SRE book: Announcing the second edition of "Site Reliability Engineering”

A decade after the original release, Google is publishing a completely rewritten second edition of their Site Reliability Engineering book. This update is necessary due to profound shifts in the computing landscape, including the rise of AI, microservices, and multi-cloud. The first edition focused on early cloud adoption and monolithic architectures, whereas the new edition addresses the challenges of maintaining reliability in complex, modern, globally scaled systems. Practices like SLOs, monitoring, automation, and reliability testing have matured significantly over the past ten years. For instance, SLOs have evolved from basic availability metrics to user-centric, multi-tiered systems with automated enforcement. Monitoring has transformed from static dashboards to leveraging OpenTelemetry and distributed tracing for quick root cause analysis.Automation has moved from custom scripts to declarative infrastructure as code and internal developer platforms. Reliability testing now includes continuous resilience checks and live production canary verification. Capacity management has shifted to dynamic auto-scaling and cloud FinOps alignment. Organizational models for SRE have also diversified beyond dedicated teams. The new edition also delves into the dual role of SRE and AI, explaining how SRE supports AI infrastructure and how AI enhances operational reliability. It covers operating large language models and using AI tools to improve incident triage and postmortem generation. Despite technological advancements, the core principles of SRE and the importance of a reliability-focused culture remain unchanged. This rewritten edition is intended as a definitive roadmap for the next era of production systems.
https://cloud.google.com/blog/products/devops-sre/sre-book-second-edition/ cloud.google.com
CdXz5zHNQW_oDLtzwcL21.jpeg
RSS Hunter RSS Hunter • Oct 12

Introducing TimesFM 3.0 with multivariate time-series forecasting in BigQuery

BigQuery is introducing TimesFM 3.0, a new foundation model for multivariate time-series forecasting directly within SQL. Previously, AI.FORECAST supported only univariate forecasting. TimesFM 3.0, developed by Google Research, excels at multivariate forecasting, ranking highly on public leaderboards like GIFT-Eval. It can forecast multiple correlated target columns simultaneously. The model also supports covariate forecasting, using both past and future known values. This enables more accurate predictions by incorporating external drivers like weather or promotions. Existing univariate forecasting and anomaly detection workflows can benefit from TimesFM 3.0 by simply updating the model specification. Multivariate forecasting in BigQuery requires a specific input table format with historical and future horizon rows. Users can also leverage natural language prompts in BigQuery Conversational Analytics for multivariate forecasting. This new capability eliminates the need for complex data pipelines and model infrastructure for enterprise-grade forecasting.
https://cloud.google.com/blog/products/data-analytics/bigquery-timesfm-3-0-with-multivariate-time-series-forecasting/ cloud.google.com
CdXz5zHNQW_DhZaejS59R.gif
RSS Hunter RSS Hunter • Oct 12

Trusted Data Apps for the Agentic Data Cloud, powered by Looker

Business intelligence users often struggle with standardized dashboards and limited chart options, requiring cumbersome CSV exports for deeper analysis. Designing custom dashboards that meet specific needs, like executive briefings or customer portals, is also challenging with typical grid-based tools. Previously, this led to building custom applications, a process now faster with AI but still hindered by production deployment issues. These issues include a lack of governed data, row-level security, and a clear security review path. To address this, Looker is launching Looker Data Apps, which combine AI's speed with Looker's governance framework.Data apps are AI-generated, interactive web applications hosted within Looker, designed for specific operational workflows. They offer bespoke layouts, allow user input for scenario testing and simulations, and connect analysis to execution tasks like updating records or dispatching events. These apps are built using AI developer tools grounded in Looker's semantic layer. They expand upon dashboards by enabling active execution and custom web components, unlike dashboards' aggregate visibility and standardized charts.Looker Data Apps are grounded in governed data, ensuring accuracy and consistency through the semantic layer. They transform analysis into action with customizable, multi-page web app experiences. Designed for enterprises, these apps inherit existing semantic models, row-level security, and access controls. They offer zero-code deployment of agent-generated applications, with no runtime token costs for end-user interactions. Deploying data apps is as simple as embedding a dashboard, allowing them to be integrated into external portals seamlessly. Looker Data Apps enable the transition from prototype to production-grade tools in minutes, bridging the gap between basic dashboards and complex custom systems.
https://cloud.google.com/blog/products/business-intelligence/introducing-looker-data-apps/ cloud.google.com
CdXz5zHNQW_y0AsX1lPrs.gif
RSS Hunter RSS Hunter • Oct 12

Agentic data science with the Data Agent Kit

Payment delays prompted a compliance team to suspect fraud, requiring a data scientist to identify transactions for investigation. This process involves ingesting and cleaning raw transaction logs, training a fraud detection model, scoring new transactions, and automating the workflow. The Data Agent Kit within the Antigravity IDE simplifies this end-to-end process.The kit offers MCP tools and Markdown skills to connect agents to Google Cloud data services and provides guidance for data stack interactions. It also enables users to manage Google's Agentic Data Cloud directly from their IDE. Initially, transaction logs are loaded into BigQuery using Spark via an agent-generated notebook.Following data loading, a dbt project is created for cleaning and enriching the BigQuery data, including handling duplicates and joining with reference information. This dbt project also includes a uniqueness test for transactions. Next, a training notebook is generated to build a Random Forest fraud detection model using historical data with known fraud labels.The trained model is then used in an inference notebook to score unlabeled transactions. Transactions with a fraud probability exceeding a specified threshold, like 50%, are sent to a Spanner table for review. This process creates a shortlist of transactions for the compliance team to investigate.Finally, Managed Service for Apache Airflow is utilized to automate the entire workflow, ensuring it runs daily. The agent generates an Airflow DAG definition and deployment configuration, orchestrating the data loading, transformation, and scoring steps. This automated pipeline continuously populates the review queue with potentially fraudulent transactions.
https://cloud.google.com/blog/products/data-analytics/bringing-you-the-data-agent-kit/ cloud.google.com
CdXz5zHNQW_pxgNhwMrDK.gif
RSS Hunter RSS Hunter • Oct 12

Modernizing Unstructured Data Workflows: Alteryx Live Query meets Google Cloud BigQuery

Alteryx One Live Query and Google Cloud BigQuery offer a powerful solution for handling complex data. This integration minimizes data movement by executing transformations natively within BigQuery, a concept called SQL pushdown. Live Query provides a no-code, browser-based interface for building data pipelines. It seamlessly orchestrates Google's AI models, like Gemini, to extract intelligence from unstructured documents such as PDFs. This approach is particularly beneficial for finance and operations teams dealing with inconsistent vendor invoices. Traditionally, invoice processing involved manual effort or fragmented tools, leading to errors and delays. The Alteryx and BigQuery integration automates this entire lifecycle, from extraction to standardization, directly within BigQuery. This ensures data governance and security by keeping sensitive information within the warehouse. The workflow begins with extracting structured fields from PDFs using AI models. Results are then standardized and compared against existing historical invoice data in BigQuery. This reconciliation process quickly identifies potential duplicates and enables faster exception handling. The system applies business rules to flag further anomalies, leading to governed operational outputs. This Google Cloud-native pattern keeps data resident in BigQuery, leveraging its security and control features. The browser serves as the authoring layer, Alteryx One Platform Services manage coordination, and Google Cloud acts as the governed data and execution layer.
https://cloud.google.com/blog/products/data-analytics/modernize-unstructured-data-workloads-with-alteryx-and-bigquery/ cloud.google.com
CdXz5zHNQW_nRdt9JT0eP.jpeg
RSS Hunter RSS Hunter • Oct 9

Innovation in Ireland: How Irish brands scale with Gemini Enterprise

Ireland has become a significant global technology hub, with organizations across sectors adopting AI. Companies are moving from AI prototypes to production-ready deployments using Gemini Enterprise. This advancement reinforces Ireland's competitive standing in Europe and aligns with its National Digital and AI Strategy. Google's local teams are working with industry leaders to implement Gemini Enterprise for various AI-powered solutions. Google's operations in Ireland contributed an estimated 10 billion euros to the GDP in 2025 through AI and cloud services. Ryanair is modernizing its operations and infrastructure for 35,000 employees, improving decision-making and productivity. Smyths Toys enhances customer support with an AI agent named Codie, resolving over 60% of web inquiries. The Irish Revenue Commissioners are improving public services by migrating to a unified AI architecture for secure data analysis. Virgin Media Ireland is accelerating workflows and deploying AI agents, leading to faster workload deployment. Numerous Irish startups are also leveraging Google Cloud and Gemini Enterprise to build competitive international services, impacting areas like sports analytics, healthcare, and carbon offsetting.
https://cloud.google.com/blog/topics/customers/ireland-innovation-companies-startups-governments-scale-with-gemini/ cloud.google.com
RSS Hunter RSS Hunter • Oct 8

Google Public Sector and SUNY launch AI-enabled platform to accelerate university research

Public research universities face the challenge of accelerating discoveries by adopting advanced technology. The emergence of agentic AI presents an opportunity to process vast datasets and complex workflows, speeding up scientific progress. Google Cloud, in collaboration with the State University of New York (SUNY), has launched the SUNY AI Platform. This platform provides a secure environment for AI and high-performance computing, initially supporting nine research-intensive SUNY campuses. Researchers can leverage Google Cloud's advanced capabilities, including storage, computing, and Gemini Enterprise, within a unified ecosystem. The platform aims to simplify data analysis, foster collaboration, and ensure responsible AI practices across institutions. Beyond technology, SUNY is equipping its community with AI-ready skills through free access to Google's self-paced AI credentials. These courses, designed for all experience levels, focus on practical skills like prompt writing and responsible AI tool usage. SUNY is also integrating Google Career Certificates into academic programs to enhance student employability. This initiative combines technological access with the necessary skills and institutional support for responsible AI implementation. The partnership aims to accelerate research with real-world impact, benefiting students, faculty, and New York's workforce.
https://cloud.google.com/blog/topics/public-sector/gps-suny-launch-ai-enabled-platform-to-accelerate-university-research/ cloud.google.com
CdXz5zHNQW_SRYhgeVHAZ.jpeg
RSS Hunter RSS Hunter • Oct 8

Welcome to Gemini at Work 2026: Introducing the Gemini agent

Google Cloud announces Gemini, a universal agent designed to streamline work processes. This agent answers questions, handles knowledge work, creates media, and writes and runs code from a single interface. It integrates directly into Google Workspace applications like Gmail and Docs, maintaining context and control across all platforms. New data and analytics skills allow both technical and business users to extract operational insights using plain language. Gemini also offers industry-specific tools for finance and legal teams, enhancing specialization. Security is paramount, with robust governance through identity, policy management, and sandboxing. Cost controls are managed through multi-model orchestration and real-time spend caps. Leading companies are already leveraging Gemini Enterprise, with nearly 90% of Fortune 100 companies using it. The agent connects to existing systems via a tools registry and learns through reusable skills and a comprehensive memory system. Gemini personalizes its assistance by learning from interactions, users, and organizational context. This unified agent aims to transform how work is initiated and executed, moving beyond experimentation to operational deployment.
https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026/ cloud.google.com
CdXz5zHNQW_Rcp9ipuVxN.png
RSS Hunter RSS Hunter • Oct 8

Empowering SMBs to do more with Gemini

Google has announced Gemini, a new universal agent designed to enhance efficiency and growth for small and midsize businesses. Gemini acts as a personal assistant and team member, integrating with existing tools and data to handle tasks from answering questions to writing code. It connects to popular SMB platforms like Asana, Salesforce, and Slack, learning and improving with use. Gemini offers model choice, allowing it to leverage the best model for each job, optimizing accuracy and cost. The agent is accessible across web, mobile, and desktop applications, ensuring widespread usability. Cost controls are integrated to manage expenses effectively, with features like multi-model orchestration and spend caps. Early access customers are already reporting significant benefits, such as reduced troubleshooting times and automated project management. Examples include Aerotech improving production capacity and Angelus streamlining R&D. EdSpan Global automates learning object standardization, while Eficode enhances customer support. KLog.co drastically cuts data entry errors and increases document processing. Koufu automates sales reporting, and La Maison du Whisky creates marketing copy. NEEOH standardizes AI usage for campaign generation, and Quadrant Health Group automates back-office tasks. Salt Productions and Pinch Creative consolidate fragmented files for easier access. Sep 2 deploys AI agents for threat monitoring, and Stratagem automates administrative workflows. Sunhouse improves access to archived design work, saving valuable time. Google also offers free training programs and a partner network to support SMB adoption of AI.
https://cloud.google.com/blog/topics/startups/how-to-grow-your-small-business-using-google-gemini/ cloud.google.com
RSS Hunter RSS Hunter • Oct 8

Introducing Google Cloud’s U4 compute: Enabling ultra-low latency trading

Capital markets demand predictable performance and low latency, facing limitations with traditional data centers. Google Cloud has launched its Ultra Low Latency (ULL) Solution to address these challenges. This solution combines the speed of co-location with cloud scalability and automation. It offers ultra-low latency networking and agility for high-frequency trading workflows in the cloud. The ULL Solution is built on three key pillars: hardware-level networking for data distribution, advanced networking observability for analysis, and deterministic high-performance compute for execution. These pillars provide capabilities like bare metal performance, high-performance storage, and physical traffic isolation. Hardware-accelerated multicast distribution bypasses traditional virtual switches for direct market data streaming. Accelerated packet processing and precision timing with UTC synchronization ensure accurate and predictable data handling. Built-in telemetry provides continuous low-latency packet capture for compliance and analytics without performance impact. The U4 machine family, featuring U4P, U4C, and U4S series, powers the ULL Solution with dedicated bare metal and high-performance VMs. This architecture enables core exchange systems and trading strategies to run in the cloud with the required speed and consistency. Storage options include fast local Titanium SSDs for logging and scalable Hyperdisk for archives. The ULL Solution is available in select Google Cloud regions, empowering financial markets with cloud benefits for today and future AI-driven trading.
https://cloud.google.com/blog/topics/financial-services/ultra-low-latency-solution-with-u4-enables-high-velocity-trading/ cloud.google.com
RSS Hunter RSS Hunter • Oct 7

Announcing MCP Toolbox Java SDK v1.0: Agentic data access for the enterprise

Google proudly announces the release of MCP Toolbox Java SDK version 1.0, enhancing agent orchestration for enterprise Java. This SDK provides type-safe agent management, crucial for mission-critical AI deployments due to Java's concurrency and state management capabilities. The Model Context Protocol (MCP) acts as a universal interface, simplifying integration by decoupling AI models from data sources. This eliminates the need for numerous custom connections, reducing security risks and maintenance overhead.Key improvements in Java SDK v1.0 include transport layer abstraction for flexible network implementations and decoupled client authentication for enhanced security compliance. Native support for default parameters reduces payload sizes, while pruning bound parameters prevents sensitive data exposure to LLMs. Standardized version selection and robust session tracking ensure protocol consistency. Developers can easily attach custom headers to outgoing requests for added metadata.A real-world example demonstrates an autonomous transit concierge application using AlloyDB and Spring Boot. This agent answers complex travel queries by integrating unstructured pet policies with structured schedules and booking functionalities. The MCP Toolbox maps natural language intents to secure SQL queries via a tools.yaml configuration.Stateful agent architecture is achieved using Spring Boot and LangChain4j, managing conversational memory within HTTP sessions. This architecture allows for clean separation of concerns, with a declarative agent interface and a tool execution service. The SDK simplifies session management, defines agent personas, and enables type-safe tool execution.Authentication is secured by default using Application Default Credentials (ADC), eliminating hard-coded keys. Runtime warnings alert developers to potential credential exposure over plaintext HTTP. The decoupled nature of MCP Toolbox and the Spring Boot agent allows independent scaling on Google Cloud Run. Developers can easily add the SDK dependency to their Maven or Gradle projects. Resources like the GitHub repository and demo application are available for getting started.
https://cloud.google.com/blog/topics/developers-practitioners/announcing-mcp-toolbox-java-sdk-v10-agentic-data-access-for-the-enterprise/ cloud.google.com
RSS Hunter RSS Hunter • Oct 6