Anthropic's Claude Mythos AI model accessed without authorization
Anthropic announced in April 2026 it was investigating unauthorized access to its Claude Mythos AI model, an advanced system capable of detecting software vulnerabilities. The access reportedly occurred through a third-party development system rather than Anthropic's own infrastructure, and the group involved appeared to lack malicious intent.
What happened
On April 21, 2026, Anthropic announced it was investigating reports of unauthorized access to its Claude Mythos AI model by some users.
The unauthorized access reportedly occurred through an external company's system used during the model's development phase, rather than through Anthropic's own infrastructure.
Anthropic stated its own systems were not compromised.
Reported impact
Mythos is capable of identifying software security vulnerabilities in short timeframes, and unauthorized access could enable malicious actors to exploit critical infrastructure.
According to media reports, the group that accessed Mythos was interested in unpublished information about AI models and had no apparent intent to use it for cyberattacks.
Index analysis
Anthropic limited Mythos access to approximately 50 authorized organizations including Apple and Microsoft to prevent malicious use.
The incident illustrates challenges in preventing diffusion of advanced AI technology despite controlled distribution efforts.
Governance controls
Anthropic restricted Mythos to limited distribution among authorized organizations and enterprises to prevent misuse.
The company has faced multiple information security incidents including leaks of Claude Code development details and Mythos capabilities prior to public announcement.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.