Anthropic is restricting access to its powerful new AI model, Claude Mythos, due to security concerns regarding its advanced ability to identify and potentially exploit critical digital infrastructure vulnerabilities.

Key Points

• Anthropic is limiting Claude Mythos access to select partners because the model can detect complex cybersecurity flaws that often evade human testers.
• Cryptocurrency firms including Coinbase, Binance, and Fireblocks are actively seeking access to the model to bolster their internal security and penetration testing efforts.
• Anthropic claims the model has successfully identified security vulnerabilities in legacy systems that remained undetected for nearly 30 years.
• OpenAI is simultaneously launching its own limited-release cybersecurity tool to compete with Anthropic’s specialized AI offerings.

Why it Matters

The high demand from cryptocurrency exchanges highlights the urgent need for advanced automated tools to protect billions of dollars in digital assets from sophisticated cyberattacks. If these AI models become widely available, they could significantly shift the balance of power between security teams defending infrastructure and malicious actors attempting to exploit it.