A security researcher claims to have achieved a prompt-based jailbreak against Anthropic's newly launched Fable 5 AI model, but the company has pushed back, asserting the technique does not represent a genuine bypass of safety controls.
The alleged exploit surfaced shortly after Fable 5's release. According to the unnamed hacker, they found a way to manipulate the model into generating restricted content. However, Anthropic released a statement disputing the finding, saying the reported method does not break the model's core safety mechanisms.
Technical details of the claimed vulnerability remain sparse. The researcher reportedly used a series of carefully crafted prompts to elicit responses that would normally be blocked. Anthropic has not disclosed what specific guardrails were allegedly overcome, making independent verification difficult.
The company emphasized that it continuously monitors for real threats and that this particular claim appears to be a misunderstanding or an ineffective attack. It urged users to rely on official channels for security disclosures.
No patches or mitigating steps have been announced, as Anthropic does not consider the issue a valid vulnerability. The incident highlights ongoing tensions between AI developers and the hacking community over what constitutes a genuine jailbreak.