Anthropic contests the order, warns a single-jailbreak standard would freeze new deployments
The directive followed a claimed jailbreak of Mythos 5. Anthropic says its review found only previously known minor vulnerabilities and that identical prompts produce comparable outputs from public models including GPT-5.5, arguing that applying this standard industry-wide would halt frontier model launches.
why it mattersIf a single demonstrated jailbreak can trigger global suspension, every lab's release calculus changes overnight.