OpenAI slows release of Astra model citing cyber capabilities

OpenAI has slowed the release of its AI model Astra after its cybersecurity capabilities reached a 'critical' threshold, indicating it could autonomously exploit zero-day vulnerabilities and execute complex cyberattacks without human help. In response, OpenAI is implementing stricter controls including isolated testing environments and collaborating with government and safety bodies to manage risks.

This Wire brief sits within Fusion42's coverage of AI Infrastructure, and 13 sources have reported it between 7 Aug 2026 and 10 Aug 2026.