Anthropic's AI Security Tool Mythos Reportedly Accessed by Unauthorized Group via Third-Party Vendor

Bloomberg reports that an unidentified private online forum has obtained access to Anthropic's exclusive cybersecurity AI tool Mythos through a third-party vendor. The group, active on a Discord channel focused on unreleased AI models, gained entry on the same day Mythos was publicly announced. They used knowledge of Anthropic's model format to locate the tool online and have been regularly using it since access was obtained, providing Bloomberg with screenshots and live demonstrations as evidence. Anthropic confirmed it is investigating reports of unauthorized access through a vendor environment but stated no evidence suggests its own systems have been impacted. The company designed Mythos as an enterprise security product that could potentially be weaponized as a hacking tool if misused. As part of Project Glasswing, Mythos was released to a limited number of vendors including Apple to prevent use by malicious actors. One group member interviewed by Bloomberg is employed at a third-party contractor working for Anthropic. The unauthorized users claim interest in experimenting with new models rather than causing harm.