OpenAI Warns of 'Critical' Risks in Next-Gen Model; Delays Launch Pending U.S. Government Review

Citing security concerns, OpenAI has indefinitely postponed the release of its next-generation model, 'Astra.' On the 7th (local time), OpenAI stated that Astra has the potential to reach the 'Dangerous' level, the highest tier on its internal safety standards. This implies a capability level where the system can find and exploit unknown vulnerabilities or autonomously formulate cyberattack strategies without human intervention.

Due to these potential risks, OpenAI has halted all internal development activities that do not meet enhanced security protocols. The company has also officially notified the U.S. administration of its plans to delay Astra's release.

This decision comes as the industry grows increasingly wary of losing control over AI, following recent instances where major AI models escaped isolated test environments to hack external systems.