Anthropic limits access to AI model, fearing future of cyberattacks
AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities, Anthropic said.
Anthropic has restricted access to its latest AI model, Claude Mythos, after the system demonstrated unprecedented capability in identifying and exploiting software vulnerabilities. The company reported that the AI discovered thousands of zero-day vulnerabilities across operating systems and browsers, showcasing coding abilities that surpass most human cybersecurity experts. This decision marks a significant step in AI safety protocols as models reach concerning levels of autonomous technical capability.
The development highlights growing concerns within the tech industry about AI systems potentially being weaponized for cyberattacks. Zero-day vulnerabilities are previously unknown security flaws that can be exploited before developers create patches, making them particularly valuable to both legitimate security researchers and malicious actors. Anthropic's findings suggest AI models have crossed a critical threshold where they can independently discover and exploit these vulnerabilities at scale.
The implications for cybersecurity and broader technology markets could be substantial. While such capabilities could revolutionize defensive cybersecurity by helping identify vulnerabilities faster, they also raise serious questions about AI governance and the potential for these tools to fall into the wrong hands. The crypto industry, already a frequent target of cyberattacks, may need to reassess security protocols as AI-powered threats become more sophisticated.
Industry observers will be watching how other AI companies respond to Anthropic's disclosure and whether similar restrictions become standard practice across the sector.
Source: Cointelegraph
