SACKS: ANTHROPIC REFUSED JAILBREAK FIX

Sacks, speaking publicly about internal Anthropic discussions, said a credible source identified a Fable jailbreak—a method to bypass the model's safety constraints. According to Sacks, Amodei declined both to fix the vulnerability and to de-deploy the affected model. Fable refers to Anthropic's newly released Mythos class models launched earlier this week. Sacks indicated he has discussed the situation with stakeholders inside and outside government. The allegation raises questions about Anthropic's security protocols and response procedures for vulnerability reports. The company has positioned itself as prioritizing AI safety, making the reported refusal to address or remove the compromised model significant within that context. Neither Anthropic nor Amodei have publicly responded to Sacks's claims. The timing coincides with increased scrutiny of large language model safety practices across the industry.