Anthropic has announced the development of a new AI model, ‘Claude Mythos Preview’, which it does not intend to release to the general public, citing serious cybersecurity risks. According to the company, the system demonstrates unprecedented capabilities in identifying vulnerabilities (bugs) in software, even in highly secure systems such as browsers and operating environments.
Unlike traditional methods that required months of specialized research, the model can detect previously unknown bugs within minutes. Notably, it reportedly identified a flaw that had remained undiscovered for 27 years in OpenBSD, an operating system known for its strong security.
Anthropic is granting access to the new model only to around 40 tech companies (e.g. Apple), aiming to help them strengthen their defenses against cyberattacks.
The company warns that such capabilities could be exploited maliciously, potentially enabling even non-experts to carry out large-scale attacks. Some experts believe this may be the first AI system capable, under certain conditions, of disrupting major corporations (e.g. those listed in the Fortune 100), crippling entire segments of the internet, or even breaching sensitive government systems.
For this reason, access to the model is restricted to approximately 40 major companies, including Apple, Amazon, Microsoft, and Google, as part of an initiative called ‘Project Glasswing’, aimed at strengthening defenses against future cyber threats.”
Πηγές: The New York Time, Axios
What is fyi.news?
