OpenAI Slows Astra Development, Citing Possible 'Critical' Cyber Capabilities
OpenAI told Axios it cannot rule out 'critical' cyber capabilities in its upcoming Astra model — the highest level of its Preparedness Framework — pausing internal work and tightening security before any release.
What happened
OpenAI told Axios it "cannot rule out critical cyber capabilities" in Astra, one of its upcoming models. Internal evaluations showed significant advances in agentic coding and cybersecurity, the most serious level under the Preparedness Framework first published in 2023.
The company is scaling up testing and security and slowing Astra's development, pausing internal work that does not meet stricter requirements. The framework demands safeguards even before release.
Tighter controls, a delayed release
According to the report, OpenAI is adding isolated test environments, universal monitoring across agentic applications, restricted network and tool access and encryption of model weights. Any release could slip further.
"OpenAI voluntarily informed the administration of their plans to delay the release," a White House official said.
Why it matters
Offensive cyber capability is now a first-class safety category for frontier labs, and speed is being traded for security. Expect more schedule uncertainty around top models and more safety gates on agentic workflows.