Rep. Andrew Garbarino, chair of the House Homeland Security Committee, emerged visibly shaken after a closed-door demonstration of Anthropic's Mythos model, the company's most advanced AI system. 

"They showed us what it could do, and it's scary," Garbarino told Punchbowl News. In the briefing, Anthropic told Mythos to find a flaw in a banking system and withdraw money from accounts, which the machine did. It then discovered and resolved the same flaw, showing both offensive and defensive abilities that have raised eyebrows among regulators and lawmakers.

However, Mythos has found many severe vulnerabilities, some of which have been zero-day exploits targeting popular operating systems and browsers.

Specifically, the automated model found a 17-year-old remote code execution flaw (CVE-2026-4747) in FreeBSD, giving full server access to attackers.

On its part, Anthropic showed how Mythos can use such an exploit chain to combine sandbox escape and privilege escalation to deliver kernel access via web pages.

The Federal Reserve organised emergency discussions with key banking CEOs in April 2026 due to the issues that were brought to light through Mythos. The governor of the Bank of England, Andrew Bailey, made an appeal through the Financial Stability Board for a meeting with Anthropic over cybersecurity vulnerabilities brought about by the model.

Project Glasswing was started by Anthropic in April 2026, allowing access only to certain US banks and tech companies, namely JPMorgan Chase, Amazon, and Microsoft.

Garbarino acknowledged the scale of the challenge. "I say act, but I don't know what the answer is," he said, adding that "95% of my colleagues don't understand what the hell's going on" regarding AI capabilities. 

He praised Trump's executive order on AI-driven cyber threats and backed a proposed federal AI regulatory framework from Representatives Jay Obernolte and Lori Trahan.