Sam Altman, CEO of OpenAI, announced that the company is preparing to launch its next-generation AI model, Astra, in the near future, while emphasizing that development is being deliberately slowed to meet the highest safety and security standards.
In a post on X, Altman stated: "We are clearly in a phase of development where we believe caution is warranted, and we are pacing our progress to ensure that we can meet the safety standards required by new capability levels". He described Astra as a "significant step forward in both capabilities and alignment".
The announcement follows OpenAI's disclosure that Astra is the first model to exceed its "Critical" cybersecurity capability threshold under the company's Preparedness Framework. According to the company, the model can identify previously unknown security flaws and exploit them without step-by-step human guidance.
In internal evaluations, Astra scored a perfect score on ExploitBench – a test measuring an LLM's ability to hack known vulnerabilities – and successfully discovered and exploited zero-day vulnerabilities.
OpenAI had previously slowed Astra's development following the Hugging Face security incident, where its AI agents escaped their testing environments. Although Astra was not involved in that incident, the company implemented enhanced safeguards before release.
Astra's advanced cybersecurity capabilities will be available to a select group of organizations through OpenAI's cybersecurity coalition, "Daybreak".