Recent testing of Anthropic’s latest artificial intelligence model, Mythos 5, demonstrated unprecedented cyberattack capabilities, according to The Economist.
Senator Mark Warner, Vice Chairman of the U.S. Senate Intelligence Committee, revealed that Mythos 5 was able to gain access to nearly all classified systems of the U.S. National Security Agency (NSA) not in weeks, but in just a matter of hours. The information was reportedly shared with lawmakers by NSA Director and Commander of U.S. Cyber Command General Joshua Rudd. The exact level of access obtained by the AI system has not been disclosed.
According to officials, the system demonstrated capabilities that would typically require years of work by experienced cybersecurity professionals.
Anthropic officially unveiled Mythos 5 on April 7, 2026. The company stated that the model is capable of identifying and exploiting critical vulnerabilities across major operating systems, web browsers, and enterprise software platforms. According to Forbes, many of the vulnerabilities discovered by Mythos 5 had existed for decades and had remained undetected despite repeated security audits conducted by cybersecurity experts.
The model was tested as part of Project Glasswing. Anthropic reported that Mythos 5 helped identify more than 10,000 high-severity and critical vulnerabilities. Company executives acknowledged that these capabilities could potentially transform the system into a powerful tool for hackers and foreign intelligence agencies. As a result, Anthropic decided against a public release and limited access to a select group of trusted partners.
At the same time, Anthropic developed Fable 5, the first publicly available version of the technologies behind Mythos. Unlike Mythos 5, Fable 5 includes built-in safeguards that block requests related to cybersecurity, biology, and chemistry. According to the company, Fable 5 significantly outperforms previous public Anthropic models in coding, information analysis, image processing, scientific research, and complex multi-step task execution.
The rapid advancement of AI capabilities has reportedly raised concerns within President Donald Trump’s administration. In April, Vice President J.D. Vance held a closed-door meeting with executives from OpenAI, Anthropic, Google, Microsoft, and SpaceX. Participants discussed next-generation AI systems capable of independently discovering software vulnerabilities and conducting large-scale cyberattacks that could disrupt banks, hospitals, and critical infrastructure.
The discussions reportedly prompted growing concern within the White House regarding its previous strategy of accelerating AI development. Officials began considering stronger oversight measures for advanced AI models.
On June 2, President Donald Trump signed an executive order establishing a voluntary framework for cooperation between the federal government and leading AI companies. Under the initiative, developers are expected to provide government agencies with access to new AI models before their public release.
On June 12, U.S. authorities introduced even stricter measures. The government ordered Anthropic to restrict access to Mythos 5 and Fable 5 for foreign nationals, including non-U.S. employees working at the company. Anthropic stated that it would comply with the decision in the interest of national security and in accordance with export-control requirements.
The restrictions reportedly affect even America’s closest intelligence-sharing allies within the Five Eyes alliance, including Australia, the United Kingdom, Canada, and New Zealand.
The Economist notes that Washington’s actions resemble U.S. policies following World War II, when American authorities sought to maintain a monopoly on nuclear technologies and restricted the transfer of research even to allied nations.
Experts warn that export controls may not fully protect advanced AI technologies. Cynthia Kaiser, a former FBI cyber division official, noted that U.S.-based credentials for accessing restricted AI services are already available on black markets.
Anthropic is a U.S.-based artificial intelligence company founded in 2021 by former OpenAI employees. Headquartered in San Francisco, the company’s founders left OpenAI due to disagreements regarding AI safety and long-term development strategy. Anthropic’s valuation has reportedly reached approximately $965 billion, making it one of the most valuable AI companies in the world.
