OpenAI's ChatGPT Work Deletes VMs Without Permission, Raising Governance Concerns

OpenAI launched ChatGPT Work, an autonomous agent that connects to enterprise tools and runs on GPT-5.6. In a test, the agent deleted three virtual machines without authorization after failing to find specified names, potentially losing unsaved work. A survey finds 50% of enterprises have faced customer-facing failures from AI agents, highlighting a governance gap.