AI SHOCKER: Anthropic's Mythos Model Exposes US Government Vulnerabilities!

A U.S. official, speaking anonymously, revealed on Tuesday that one of Anthropic's artificial intelligence models successfully identified vulnerabilities in highly sensitive and secure U.S. government computer systems during a recent testing exercise. This collaboration involved Anthropic teaming up with U.S. intelligence agencies, utilizing the company's advanced Mythos model. The official noted that the AI model pinpointed specific vulnerabilities within mere hours, though they stressed this does not imply the model was capable of exploiting these flaws within the same timeframe.

This critical testing was conducted under Anthropic's initiative known as Project Glasswing. This project brought together leading technology companies and other organizations with the overarching goal of fortifying the world's essential software against severe potential fallout that could arise from advanced AI models like Mythos, which pose risks to public safety, national security, and the global economy.

The testing was briefly mentioned by Democratic Sen. Mark Warner of Virginia during a Senate Committee on Banking, Housing, and Urban Affairs hearing on June 11. Senator Warner sensationally stated, "This tool broke into almost all of our classified systems, not in weeks but in hours," attributing this information to Gen. Joshua Rudd, who heads the National Security Agency and U.S. Cyber Command. Both the NSA and Anthropic declined to comment on these specific matters when contacted.

Despite this recent cooperation to bolster cybersecurity, tensions have been escalating between the California-based AI company and the Trump administration. Anthropic has previously voiced concerns regarding the potential use of its AI by the U.S. military, while the administration has, in turn, imposed restrictions on the deployment of some of Anthropic's models.

Earlier this month, the administration issued a significant directive compelling Anthropic to prevent foreign nationals from accessing and utilizing its latest artificial intelligence models, specifically Fable 5 and Mythos 5. Fable 5, a more limited version of the highly advanced Mythos model, was released widely by Anthropic earlier this month. Access to the more powerful Mythos model has been tightly restricted by the company itself due to significant cybersecurity fears.

This governmental directive followed closely on the heels of President Donald Trump's executive order, signed ten days prior. The order established a framework for the federal government to rigorously vet the national security risks associated with the most advanced AI systems for up to a month before their public release, though participation by AI developers was stated to be voluntary.

In response to the administration's mandate, Anthropic announced it had disabled these models for all of its customers to ensure compliance. However, the AI giant also expressed its belief that the measures taken by the government were not warranted by the specific security concern it had initially flagged. The company's stance highlights a growing debate over the appropriate level of government oversight and control regarding cutting-edge AI technologies.

Further complicating the issue, a coalition of over 100 cybersecurity executives and leaders from prominent companies, including Adobe and Nvidia, have formally petitioned the Trump administration to revoke its directive. In a letter to the government, they argued that such a move could inadvertently benefit U.S. adversaries more than it safeguards national interests. While acknowledging that Anthropic's Mythos models are "quite good" at discovering software flaws and weaponizing exploits, the signatories stressed they are "not uniquely good at these tasks." Many experts who signed the letter indicated their regular use of other foundational and open-source models for security audits and training. The letter concluded by warning against removing America's best cyber defense capabilities "without a good reason," particularly at a time when adversaries are making rapid advancements in the cyber domain.