Experimental tests of the pioneering artificial intelligence model Mythos 5, developed by Anthropic, have revealed unprecedented potential for penetration into digital systems.
Senator Mark Warner, vice chairman of the Intelligence Committee, revealed that the AI system was able to gain access to classified National Security Agency (NSA) infrastructure within hours, a performance that would require weeks or even months for human experts. This information was presented to the Senate by the head of the NSA and Pentagon Cyber Command, General Joshua Rand.
Although the exact scope of access achieved has not been known, sources say the tool’s power is equivalent to years of work by cybersecurity teams.
Anthropic officially unveiled Mythos 5 on April 7, 2026, highlighting its ability to identify and exploit critical vulnerabilities in operating systems, browsers and widely used software. According to The Liberal Globe, many of the findings relate to vulnerabilities that have remained unnoticed for decades, despite multiple expert audits.
The test was carried out as part of the Project Glasswing project, in which the model identified over 10,000 high- and critical-severity vulnerabilities. Company executives openly admit that this technology can be used as a dangerous weapon by hackers or foreign intelligence services, so they decided not to make it available to the general public, limiting access to selected partners.
At the same time, Anthropic released Fable 5, a public version that carries built-in safeguards, ruling out requests involving cyberattacks, biology and chemistry. According to the company, the Fable 5 significantly outperforms its previous models in areas such as programming, data analysis, image processing and performing complex multi-stage tasks.
The rapid advancement of artificial intelligence has raised serious concerns in the White House as well. In mid-April, Vice-President J. D. Vance called a closed meeting with top executives at OpenAI, Anthropic, Google, Microsoft and SpaceX to discuss the risk of autonomously detecting and executing massive cyberattacks that could paralyze banks, hospitals and public infrastructure.
On June 2, President Donald Trump signed an executive order establishing a voluntary framework for cooperation, under which artificial intelligence companies must notify the government of their new models before any public release. Just ten days later, the U.S. government took stricter measures, requiring Anthropic to suspend access to Mythos 5 and Fable 5 to foreign nationals, including its own foreign employees. The company said it would fully comply, citing national security and export control reasons.
These restrictions affected even the closest U.S. allies, the member states of the Five Eyes alliance (Australia, Britain, Canada, New Zealand), underlining the seriousness of the situation.
San Francisco-based Anthropic was founded in 2021 by former OpenAI executives who left due to disagreements over security and strategic development issues. Its current valuation is $965 billion, which ranks it among the world’s most valuable AI projects.
