OpenAI outlines restricted access plans for forthcoming Astra model
OpenAI said its forthcoming Astra model can autonomously find and exploit software flaws and that access to its most advanced cybersecurity capabilities will be limited.
OpenAI described Astra as its first model to meet the company's critical cybersecurity threshold and said it plans to release the model soon. The company reported that Astra exploited two previously unknown flaws in a modified evaluation.
The company said it will use account risk controls, abuse detection and chain-of-thought monitoring, but TechCrunch noted that the claims and safeguards have not been independently confirmed.