Anthropic Limits AI Model Access, Citing Cyberattack Fears

Anthropic has limited access to its new AI model, Claude Mythos Preview, due to concerns over its potential use in cyberattacks. The model has demonstrated the ability to find and exploit software vulnerabilities, surpassing human capabilities in some cases. Anthropic has announced Project Glasswing, a initiative to use the model's capabilities to defensively find bugs and patch critical vulnerabilities before they can be exploited by bad actors. The move highlights the growing risks associated with AI-powered cyberattacks, which have increased by 72% year-over-year, with 87% of global organizations experiencing AI-enabled cyberattacks in 2025. Anthropic's decision to restrict access to its model underscores the need for responsible AI development and deployment, particularly in areas where it can be used to exploit vulnerabilities. A key event to watch is the progress of Project Glasswing, which is set to bring together over 40 companies, including Amazon Web Services, Apple, and Google, to combat AI-powered cyberattacks. The success of this initiative will depend on its ability to stay ahead of emerging threats and patch critical vulnerabilities before they can be exploited, with the first quarterly update on the project's progress expected in Q3 2026.