AI companies fail existential safety test as capabilities race ahead

A sweeping safety review shows leading AI companies advancing toward superhuman systems without guardrails to stop catastrophic failure