Anthropic’s Mythos AI Finds Vulnerabilities in Classified U.S. Systems During Project Glasswing

Anthropic’s Mythos is at the centre of a national security storm on June 24, 2026. The company’s Mythos model identified vulnerabilities in highly sensitive US government computer systems during an authorized testing exercise under Project Glasswing, a restricted program designed to find and fix critical software flaws before attackers could exploit them.

Senator Mark Warner, citing NSA and US Cyber Command chief Gen. Joshua Rudd, said Mythos breached nearly all classified systems in hours, not weeks. The disclosure has sent shockwaves across the AI and cybersecurity sectors.

What Project Glasswing Revealed

Anthropic refused to release Mythos publicly, instead granting access to roughly 200 vetted partners under Project Glasswing, with core participants including Amazon (NASDAQ: AMZN), Apple (NASDAQ: AAPL), Google (NASDAQ: GOOGL), Microsoft (NASDAQ: MSFT), Nvidia (NASDAQ: NVDA), JPMorgan, and the Linux Foundation. The project’s explicit goal was defensive: surface vulnerabilities before bad actors could. Anthropic’s Mythos Preview has already identified thousands of zero-day vulnerabilities, including a 27-year-old flaw in OpenBSD, one of the world’s most security-hardened operating systems.

The Senate Hearing That Changed Everything

Senator Mark Warner, vice-chairman of the Senate Intelligence Committee, delivered the key disclosure on June 11, 2026, attributing the classified breach assessment directly to Gen. Joshua Rudd, who was confirmed to lead both the NSA and US Cyber Command on March 10, 2026, by a Senate vote of 71 to 29. A US official confirmed to the AP that while Anthropic’s Mythos identified vulnerabilities within hours, identifying them did not mean the model actively exploited them within that timeframe. That distinction is critical to understanding the scope of what occurred.

Government Response and Model Restrictions

The Trump administration issued a directive restricting foreign nationals from accessing Anthropic’s newest models, Mythos 5 and Fable 5, citing national security concerns. Anthropic disabled both models for all global customers to comply with the directive, despite disagreeing that the government’s steps were warranted by the concern raised. The move drew immediate pushback from the private sector and the broader cybersecurity community.

Industry Pushback and the 100-Expert Letter

More than 100 cybersecurity experts and leaders from companies including Adobe (NASDAQ: ADBE) and Nvidia wrote to the government arguing that Anthropic’s Mythos models are capable of finding software flaws and weaponizing exploits but are not uniquely superior to other existing tools.

The letter warned it is dangerous to remove America’s best cyber defense capabilities without a compelling reason, especially as adversaries rapidly advance their own AI programs. The industry has drawn a clear line against unilateral government intervention.

Conclusion

Anthropic’s Mythos has rewritten what AI-powered cybersecurity looks like, and the implications are profound. Project Glasswing’s findings, disclosed publicly on June 24, 2026, confirm that frontier AI models now operate at a speed and depth no human team can match.

The government’s restriction of Mythos 5 and Fable 5, combined with over 100 industry signatories pushing back, frames a defining tension between national security control and AI-driven cyber defense in 2026.

Disclaimer:

The content shared by Meyka AI PTY LTD is solely for research and informational purposes. Meyka is not a financial advisory service, and the information provided should not be considered investment or trading advice