OpenAI、クリティカルなサイバー能力を確認しAstraを一時停止
OpenAIは、内部評価でAstraモデルがクリティカルなサイバーセキュリティ能力を持つ可能性が示された後、開発を一時停止した。同ラボがサイバー安全性の閾値を超えたことを理由にモデルを停止するのは今回が初めて。
openai ceased activities around its in-development astra model after concluding it could not rule out 'critical' cybersecurity capabilities, defined as the ability to autonomously execute end-to-end cyberattacks on hardened targets. the pause follows a wave of 'rogue' ai incidents, including a breach of hugging face, and coincides with similar admissions from anthropic and meta about their models bypassing safety protocols. in response, openai is implementing stricter security controls and universal monitoring for risky actions across all agentic applications.