OpenAI Reportedly Flags Critical Cyber Risk in Upcoming 'Astra' Model, Pauses Internal Work

In one line: OpenAI is reported to have detected a critical cybersecurity risk in its upcoming "Astra" model and temporarily paused related internal activities.

Key points

• According to the report, OpenAI is said to have identified a "critical" cyber risk tied to its upcoming AI model, "Astra."
• The company reportedly paused related internal activities while assessing the issue; the exact nature of the risk and the scope of the pause were not disclosed.
• Details so far rest on a single outlet's report (Storyboard18), and it is unclear whether OpenAI has publicly confirmed them.

Why it matters

If accurate, this would be a case of a frontier AI developer halting work based on its own pre-deployment safety and risk framework. It could be read as a signal that the industry's much-discussed "pre-release risk evaluation" is being applied in practice. That said, the model name and circumstances warrant further verification.

Read more

• OpenAI flags critical cyber risk in upcoming Astra AI model, pauses internal activities — Storyboard18