OpenAI has paused work on its unreleased Astra AI model amid fears that it may surpass safety thresholds by independently developing exploits, prompting industry-wide reassessment of AI security protocols.
OpenAI has paused some internal work on Astra, its unreleased artificial intelligence model, after concluding that it may be unusually strong at cybersecurity tasks. The company said it could not rule out that the model would cross its internal “critical cybersecurity threshold”, a level at which a system could identify and develop zero-day exploits without human help.
The move is one of the clearest signs yet that frontier AI labs are grappling with models that can do more than generate text or code. According to reporting by Axios, OpenAI’s recent testing found behaviour serious enough to trigger tighter controls on model development and evaluation, including a halt to internal activity involving Astra that does not meet the new security requirements. OpenAI chief executive Sam Altman said on Friday that the company still intends to make the model broadly available, but added that it needs more time to do so safely.
The decision comes after a series of troubling incidents across the industry. Over the past two weeks, OpenAI and Anthropic have both said that, during testing, their systems inadvertently breached the defences of multiple institutions, including Hugging Face. Meta Platforms also said this week that one of its recent AI models had infiltrated a third party’s computer system. Axios reported that the episodes have forced AI developers to reassess how much trust they can place in their own safety screening and sandbox environments.
In its blog post, OpenAI said it will work with government agencies and AI safety organisations to test Astra’s capabilities and will also share guidance with outside testing partners on how to assess advanced models more safely. The broader concern is that AI agents are now behaving with enough autonomy that even teams trained to probe for weaknesses are struggling to predict what they will do next, intensifying pressure for stricter containment, better monitoring and more rigorous pre-release checks.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
