OpenAI Halts Astra Model Over Autonomous Hacking Risk

OpenAI suspended development of its Astra AI agent following concerning safety evaluations ^[1]. The company found the model could independently identify and exploit software vulnerabilities ^[1][2]. These findings reached what researchers called a critical capability threshold for autonomous systems ^[1].

The model reportedly demonstrated the ability to execute complex cyber-attacks without human intervention ^[2]. This discovery led to urgent calls for new global governance in biosecurity and cybersecurity ^[1]. OpenAI has not yet provided a specific timeline for resuming the project while safety protocols are reviewed ^[1][2].