Anthropic launched two Mythos-class AI models on June 9, 2026: Claude Fable 5, available to the public, and Claude Mythos 5, restricted to vetted cybersecurity partners. Both models operate on the same underlying architecture and pricing but differ in access and safety features. The company imposed a mandatory 30-day data-retention policy on all traffic for both models, overriding previous zero-retention agreements, according to medianama.com.
Claude Fable 5 is accessible to anyone but includes classifiers that intercept high-risk prompts, redirecting them to the less capable Claude Opus 4.8 model. Claude Mythos 5, by contrast, is limited to Project Glasswing cyber defenders and critical infrastructure partners, with cybersecurity safeguards removed. Access depends on user identity rather than payment tier. Anthropic will retain all data for 30 days, using it solely to defend against novel attacks and reduce false positives, while logging human access and deleting data after the retention period in most cases, medianama.com reported.
The introduction of mandatory data retention for frontier AI models marks a potential industry precedent, especially concerning safety and privacy. This policy could influence how enterprises, including those in India, engage with advanced AI services. The dual-model approach balances broad public availability with restricted, high-risk capabilities for critical users, reflecting growing concerns about AI misuse and cybersecurity threats, according to medianama.com.
Anthropic’s data retention policy and model access controls took effect with the June 9 launch. The company stated that retained data will not be used for training models but only for security purposes, with deletion occurring after 30 days in almost all cases, per medianama.com.