OpenAI says an upcoming model may reach critical cyber capability
OpenAI says preliminary tests of its upcoming Astra model show enough progress in agentic coding and cybersecurity that it cannot rule out the highest “Critical” capability level in its safety framework. The company says Astra was not involved in the Hugging Face exploit and is adding chain-of-thought monitoring and interruption controls. For builders, the gap between a useful coding agent and an offensive cyber system is narrowing fast.

