Anthropic Investigates Unauthorized Mythos Use
Anthropic's unreleased 'Mythos' model — withheld over cyberattack risks — may have been accessed by unauthorized users.
Why it matters
A potential security breach of a deliberately restricted AI model raises immediate questions about Anthropic's access controls, the real-world risks of advanced model theft, and whether safety-by-withholding is a viable governance strategy.
The key facts
10 to knowAnthropic investigating unauthorized access to unreleased models including Mythos
Mythos deliberately withheld by Anthropic due to cyberattack capability concerns
Breach involves multiple unauthorized users
Story reported by Bloomberg via The Information
Raises governance and safety control questions
Anthropic investigating unauthorized access to unreleased models
Mythos model affected — deliberately withheld due to cyberattack capability
Unknown number of unauthorized users gained access
Published April 22, 2026 — timing suggests active/ongoing investigation
UNVERIFIED: No independent confirmation of breach scope or user identities cited
Go to the source
The Informationtheinformation.com
Publisher excerpt: Anthropic is investigating a report that a group of unauthorized users gained access to several unreleased AI models including Mythos, the model that Anthropic has withheld because of its capacity for cyberattacks, according to a spokesperson for the company. Bloomberg reported Tuesday that the ...