Hackers allegedly used Anthropic's Mythos AI to breach GitHub, steal Copilot source code

SlowMist, a blockchain security team, has disclosed that a recent leak on a cybercrime forum suggests attackers used Anthropic's Mythos — a safety-focused AI model — to breach GitHub's internal defenses. The stolen data includes roughly 4,000 GitHub internal repositories, containing Copilot source code, CodeQL algorithms, Actions runtime, and billing system code. SlowMist warns that if hackers deeply analyze this code, it could have downstream security implications for the open-source community. The reason is straightforward: these systems underpin core workflows for countless developers.