Anthropic investigating claim of unauthorised access to Mythos AI tool

Anthropic is investigating claims that a small group of individuals gained unauthorized access to its Claude Mythos AI model, a cybersecurity tool considered too powerful for public release. The company stated the breach likely occurred through misuse of access within a third-party vendor environment rather than a traditional hack. While there is no evidence that malicious actors have exploited the model or that Anthropic’s systems have been compromised, the incident raises concerns about the security of advanced AI tools. Claude Mythos has been selectively shared with certain technology and financial firms to help identify and protect against cybersecurity vulnerabilities. However, this approach depends heavily on strict access controls by those organizations. According to reports, the individuals who accessed the model already had permission to view Anthropic’s AI through work with a third-party contractor, but their use of the model outside intended boundaries poses risks. Experts warn that unauthorized use of such powerful AI could facilitate fraud, cyber abuse, or other malicious activities if the technology spreads beyond controlled environments. The incident comes amid broader discussions on the dual nature of AI in cybersecurity. At the UK’s CyberUK conference, Richard Horne, head of the National Cyber Security Centre, emphasized that while AI can accelerate the discovery and exploitation of vulnerabilities, it also offers opportunities to enhance security if properly managed. He urged organizations to focus on fundamental cybersecurity practices, such as updating software and addressing legacy IT systems, to mitigate emerging threats. Security Minister Dan Jarvis also called on AI developers to collaborate with governments to ensure responsible AI deployment. The situation highlights ongoing challenges in balancing innovation and risk management as AI tools become increasingly sophisticated. Ensuring robust oversight and secure access to advanced models like Claude Mythos will be critical to preventing misuse while harnessing AI’s potential to improve cybersecurity defenses.