OpenAI has paused development on parts of its next-generation AI model, Astra, after internal reviews raised cybersecurity concerns.
Internal review flags agentic coding risks
The company said an evaluation of Astra over the past several days found the model had made “significant advancements” in agentic coding and cybersecurity—enough to trigger OpenAI’s Preparedness Framework. That framework is designed to assess risks before a model is deployed.
According to a weekend blog post, OpenAI concluded it “cannot rule out” that Astra had reached a “critical capability level” in cyber operations. The model was not involved in any real-world incidents, including a recent autonomous cyberattack reported on Hugging Face, the company clarified.
OpenAI said it shared the findings publicly because it believes transparency with the public and the safety and security communities is important.
Competitive tensions surface in AI safety debate
The announcement arrives amid an ongoing rivalry between OpenAI and Anthropic, its closest competitor in advanced AI development. In April, OpenAI CEO Sam Altman criticized Anthropic for what he called “fear-based marketing” after the company disclosed its own model had exhibited unexpected agentic behavior.
Related: Unlocking the Benefits of Platform as a Service (PaaS) for Modern Businesses
Anthropic’s disclosure came shortly after OpenAI faced scrutiny for an autonomous cyberattack on Hugging Face, which the company later acknowledged. Some observers now question whether OpenAI’s transparency about Astra is a strategic move to shift the narrative—or simply a routine safety precaution.
Companies often delay product rollouts over potential risks, but few announce such decisions publicly while a model is still in development. OpenAI’s blog post did not specify how long the pause might last or what benchmarks Astra would need to meet before work resumes.
OpenAI has not commented on whether the decision reflects a broader shift in its development philosophy. Altman has previously spoken about the need to slow AI progress to ensure safety, though critics have questioned whether such statements are sincere or part of a competitive strategy.
For now, the company is framing the pause as a responsible step. “We are sharing this because we believe it’s important to be transparent,” the blog post stated. It remains unclear whether the move will ease concerns or fuel further debate over the pace of AI advancement.
The decision leaves Astra’s timeline uncertain. OpenAI has not disclosed when the model might be released, or whether it will undergo additional testing beyond the current internal review process.
