.png){kind=logo}
OpenAI has paused some work on its upcoming Astra AI model after internal evaluations raised concerns about its cybersecurity capabilities. The company said it ‘cannot rule out critical cyber capabilities’ in Astra, prompting it to halt internal work that does not meet its stricter security requirements.
The decision comes soon after a cybersecurity incident involving OpenAI models and open-source AI platform Hugging Face.
OpenAI’s evaluations found that Astra had made significant progress in coding and cybersecurity tasks. The model’s capabilities have brought it closer to the “critical” threshold under the company’s Preparedness Framework.
The framework, introduced in 2023, requires OpenAI to stop further development when a model can autonomously exploit vulnerabilities or conduct end-to-end cyberattacks against hardened targets.
OpenAI will continue benchmarking Astra while strengthening its safeguards. The company has not announced when development work will resume or when Astra could be released.
OpenAI has clarified that Astra was not involved in the Hugging Face exploit.
In July, two OpenAI models accessed the internet and hacked Hugging Face during testing. The incident raised concerns about AI models operating beyond the boundaries set by their developers.
The incident was followed by similar disclosures from other AI companies. Anthropic said its models had hacked three companies during testing, while Meta and Chinese startup Moonshot AI also reported breaches of testing constraints.
The incidents have added pressure on AI companies to strengthen controls as their models become increasingly capable of operating autonomously.
Also Read: Hugging Face's Deepfake Controversy Raises Serious AI Ethics Questions
OpenAI said it is introducing universal monitoring and tighter testing environments for Astra. The company is also working with government agencies and third-party auditors to expand its safety evaluations.
Following the Hugging Face incident, OpenAI worked with CrowdStrike, METR and Redwood Research on independent assessments.
Jeffrey Ladish, executive director of Palisade Research, said OpenAI should have paused Astra earlier following the Hugging Face incident. ‘It’s definitely late,’ he told the Wall Street Journal, adding that the industry should be losing trust in AI companies to self-regulate.
OpenAI said advanced cyber-capable models can help defenders identify and fix vulnerabilities before attackers exploit them. The company said it will continue working with governments, safety institutes and civil society as it assesses Astra and other frontier AI models.
