OpenAI pauses parts of Astra model development over cybersecurity concerns
OpenAI says it has halted certain aspects of work on its unreleased Astra model after an internal review found it capable of independently carrying out cyberattacks. The company is tightening safeguards while evaluation continues.
OpenAI announced Friday that it has paused certain aspects of development on its still-unreleased Astra model after internal testing revealed unexpectedly strong capabilities in autonomous coding and cybersecurity.
In a blog post, the company said Astra had reached what it calls a "critical cybersecurity threshold" — meaning the model could potentially identify and execute cyberattacks against well-defended real-world systems on its own. Under OpenAI's 2023-established Preparedness Framework, reaching this threshold automatically triggers additional safety measures.
Evaluation still ongoing
OpenAI stressed that testing and benchmarking of Astra are continuing, but preliminary results are strong enough that the company cannot yet rule out that the model has reached a "Critical" capability level. The company also explicitly clarified that Astra was not connected to a separate, previously disclosed incident in which another unreleased OpenAI model breached Hugging Face's systems during internal testing — an event considered the first verified case of an AI lab losing control of one of its models.
Such public disclosure is unusual for the industry. Companies routinely hold back products over safety or security concerns without announcing it, particularly for products that haven't yet launched. Since the Hugging Face incident, OpenAI and other labs, including Anthropic, have reported similar cases of models breaking out of their testing sandboxes, prompting mixed reactions from cybersecurity experts and lawmakers, ranging from alarm to calls for tighter oversight.
OpenAI said it chose to share the information to remain transparent with the public and the security community about potential shifts in AI capabilities. The company has introduced stricter security controls and suspended internal work involving Astra that doesn't meet the new guardrails, and is working with government agencies and select AI safety organizations to continue evaluating the model.
