OpenAI plans to make its Astra AI model broadly available, but the company is holding back the release while it addresses the model's advanced cybersecurity capabilities, CEO Sam Altman said Friday.
Altman said in a post on X that Astra is a "powerful model" and that OpenAI is working to make it "generally available." He rejected the idea of keeping highly capable AI systems restricted to a small group, writing: "We do not think it is a good strategy to keep powerful models to a chosen few."
However, he indicated that Astra's cyber capabilities require additional preparation before wider access. "Given its cyber capabilities, we need a little big longer to do this safely," Altman said, adding, "but hopefully not too long!"
OpenAI slows Astra release after internal review
OpenAI slowed development of Astra after internal evaluations raised concerns that the model could possess advanced cybersecurity capabilities requiring extra safeguards. The company said it "cannot rule out critical cyber capabilities" in Astra, leading to expanded testing, stronger monitoring, and more isolated evaluation environments.
OpenAI said the model was not linked to recent cybersecurity incidents involving other AI systems. The release timeline remains unclear as the company prioritizes safety over quickly deploying the frontier model. The decision comes as governments and AI companies face growing pressure to establish stronger standards for evaluating and deploying increasingly capable systems.
Broader AI security concerns mount
The delay follows a separate security incident involving Meta's AI systems. Meta launched an investigation, while Irregular said the incident stemmed from an evaluation error, not a sophisticated cyberattack or sandbox escape.
The Five Eyes intelligence alliance has warned that more advanced AI could rapidly accelerate cyberattacks by helping criminals discover vulnerabilities, create malicious code, and automate complex operations. The alliance said organizations may have only months to prepare for that scenario.
For executives, the takeaway is twofold. First, expect delays in frontier AI releases as vendors weigh capability against safety - plan around uncertainty rather than fixed product timelines. Second, the cyber risks that are slowing Astra are the same risks your organization will face when adopting advanced AI tools. That makes security evaluation a core part of procurement, not an afterthought. Professionals tracking these developments can follow AI for Executives & Strategy for strategic guidance, while teams responsible for defense should review AI for Cybersecurity Analysts training to prepare for the threat landscape these models introduce.
Your membership also unlocks:
