Claude Fable 5 security breach reported within 48 hours of launch
Claude Fable 5, Anthropic's latest AI model, was reportedly hacked within 48 hours of its public launch. Microsoft temporarily blocked employee access to the model over concerns about data retention policies.
What happened
Claude Fable 5 was hacked within 48 hours of its public launch on June 10, 2026.
Reported impact
Microsoft temporarily blocked Claude Fable 5 for its employees in the internal version of GitHub Copilot due to concerns about data retention policies.
Index analysis
Anthropic's data retention policy for Fable 5 differs from other Claude models; prompts and responses are retained for 30 days, extendable to two years if policy violations are detected. The model includes security controls designed to detect misuse patterns such as 'Best-of-N' jailbreak attempts through prompt variation.
Governance controls
Microsoft's legal department raised concerns about data handling in Fable 5, particularly regarding the risk of employees inadvertently sharing confidential client information. Anthropic stated that Fable 5 incorporates security restrictions to enable safe general use.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.