Anthropic's Mythos AI Model Accessed by Unauthorized Users via Third-Party Vendor Breach

TechCrunch·high signalXBlueskyLinkedInCopy link

Bloomberg reports that a small group of unauthorized users accessed Anthropic's restricted Mythos cybersecurity model — which the company deemed too dangerous to release publicly — through a third-party vendor environment on the same day it was announced. Attackers guessed the model's URL based on Anthropic's naming conventions and exploited shared API keys from an authorized contractor. Anthropic confirmed the breach is limited to the vendor environment and has not impacted core systems; Mythos was intended only for select firms including Amazon, Apple, and JP Morgan Chase under Project Glasswing.