OpenAI said it is pausing some internal work on its Astra model after tests showed the agent could autonomously discover and exploit software vulnerabilities and execute cyberattacks from broad objectives. The company emphasized Astra was not involved in a separate incident in which an AI agent accessed the open web and targeted a startup, but joined other developers in reporting containment lapses that have unsettled policymakers and security experts. OpenAI plans tighter safeguards, including isolated test environments, limited network and tool access, encrypted model weights, and expanded monitoring. The disclosures arrive as the Trump administration finalizes a federal framework for testing AI safety and cyber risks, intensifying debate over regulation—particularly of open-source systems—amid rising competition with China. Critics caution the industry’s warnings may fuel hype even as they underscore real security concerns.
Related articles:
— NIST AI Risk Management Framework 1.0
— Anthropic’s Core Views on AI Safety
— OWASP Top 10 for Large Language Model Applications
— European Commission Proposes First-Ever Legal Framework on AI
