Anthropic, a leading AI company based in San Francisco, has announced “Claude Mythos Preview”, an AI model it claims is too dangerous to release publicly due to its exceptional cybersecurity and cyber-attacking capabilities. The company says Mythos has found vulnerabilities in every major browser and operating system, potentially enabling hackers to disrupt critical software worldwide.
Security experts have described the situation as “Y2K-level alarming”. Mythos has already discovered a 27-year-old bug in critical security infrastructure and multiple vulnerabilities in the Linux kernel, which is essential for computer systems globally. These weaknesses could threaten everything from streaming services to banking systems.
If such technology becomes widely available, the implications could be catastrophic. Cyber-attacks have already crippled airports, hospitals, and transport networks, but until now such attacks required serious expertise. Mythos could put that capability within reach of amateurs and turbocharge professionals’ ability to cause havoc.
Rather than releasing Mythos publicly, Anthropic is first offering it to companies running critical infrastructure, including Apple, Microsoft, and Google, hoping they can patch security gaps before bad actors obtain similar capabilities. However, a lack of regulation means other companies may release similar models within months.
The Trump administration has declared hostility towards Anthropic, banning government agencies from using its technology and calling it a “radical left, woke company”. This animosity makes it unlikely the government will work with Anthropic to secure its own systems, which are notoriously vulnerable.
Despite some optimism—such as Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convening Wall Street executives to prepare for risks—the overall picture remains bleak. The race to secure critical infrastructure against AI-powered attacks is on, and time may be running out.
