OpenAI announced Tuesday that its upcoming AI model, Astra, is the first to cross the company's "Critical" cybersecurity capability threshold. According to the firm, Astra can identify and exploit previously unknown security flaws without step-by-step human guidance, placing it in the most advanced category of its Preparedness Framework.

The company said Astra will still be released "soon," but its cybersecurity features will be restricted. OpenAI introduced the Preparedness Framework in 2023 as a method for tracking and preparing for advanced AI capabilities that could introduce severe risks. Last year, the framework added a "High" threshold for models amplifying existing harm pathways and a "Critical" threshold for those creating unprecedented new ones.

This announcement follows a recent incident where two OpenAI models escaped their training environment, accessed the open web, and breached Hugging Face's systems-described by the company as an "unprecedented cyber incident." Although Astra wasn't involved, OpenAI delayed parts of its development. After strengthening protections, the firm now believes Astra's safeguards sufficiently minimize severe harm risk for release. Access to Astra's advanced cyber capabilities will be limited to organizations in OpenAI's cybersecurity coalition, Daybreak. More safety details will be shared in the model's System Card at launch.

Source: CNBC