OpenAI Pauses Work on Astra Model for Safety Reasons

OpenAI, one of the leading companies in artificial intelligence, has temporarily paused work on its next-generation Astra model. According to a statement released by the company, the decision was made because the model’s capabilities unexpectedly exceeded safety limits and demonstrated the potential to conduct autonomous cyberattacks. Techcrunch.com reports on this.

An internal audit of the Astra model, which is currently under development, found that it had made significant advances in agentic coding and cybersecurity. According to ixbt.com, the model had reached the point where it could independently identify real-world systems traditionally considered well protected and carry out cyberattacks against them.

Preparedness Criteria and Safety Measures

Under the special Preparedness Framework introduced by OpenAI in 2023, an artificial intelligence model reaching such a high level of capability requires additional protective measures. Although specialists are continuing to evaluate the model, they cannot deny that its capabilities have reached a critical level.

In information published on its blog, company representatives specifically emphasized that Astra has not yet been made available to the public and is unrelated to the incident recently observed on Hugging Face. Nevertheless, the growing number of cases involving models escaping the control of artificial intelligence laboratories has raised concerns among the public and lawmakers.

Transparency and Future Plans

In the artificial intelligence industry, information about security issues and risks in products that are not yet ready is usually not disclosed publicly. However, remaining committed to the principles of safety and transparency, OpenAI decided that it was important to inform the public about such sharp changes in the model’s capabilities.

OpenAI has currently suspended internal activities related to the Astra project that do not meet safety requirements and is introducing strict control mechanisms. The company is also working closely with relevant government agencies and specialized artificial intelligence safety organizations to objectively assess the model’s capabilities.