IC Oversight & Policy — 2026-06-30

House Homeland Security Chairman Garbarino Urges AI Cybersecurity Action After Mythos Demonstration to Congress Shows Real-Time Bank Vulnerability Exploitation

BLUFCongressional alarm over AI-enabled cyber offense will likely produce a formal hearing by year-end 2026, but pervasive lawmaker ignorance caps near-term federal response at executive action while China operationalizes the same capability.

Anthropic demonstrated its unreleased Mythos model to the House Homeland Security Committee in a closed session on May 13-14, showing it identify a bank vulnerability, drain accounts, and then patch the flaw 123. Chairman Andrew Garbarino (R-NY) disclosed the demonstration publicly at Punchbowl's Fly Out Day on June 28, calling it "scary" and urging Congress to act soon 24. Garbarino conceded that "95 percent of my colleagues don't understand what the hell's going on" and that lawmakers have no concrete answer on how to respond 24. Anthropic has withheld Mythos from public release; CryptoBriefing reported the model has uncovered thousands of high-severity vulnerabilities including zero-days, with limited access channeled through Project Glasswing to partners including JPMorgan Chase and Apple 3.

Analysis
Garbarino's disclosure converts closed-door alarm into political accountability and builds the committee record that drives formal hearings, making one on AI-enabled cybersecurity threats likely by year-end 2026. Moderate confidence rests on the committee's active hearing cadence and the chair's stated urgency, bounded by absent legislative consensus. Punchbowl holds the single original report. The committee press release corroborates officially, not independently. The alarm may instead serve Anthropic's procurement interests more than it reflects demand for immediate statutory action. China's Qihoo 360 already routes confirmed zero-days to Beijing within 48 hours. Glasswing and Trump's executive order remain the ceiling of federal response. A formal hearing by year-end accelerates statutory authority for frontier AI acquisition by federal agencies, forcing vendors and agency CISOs to finalize voluntary-access terms before Congress imposes them.

4 sources

Chairman Garbarino Joins Punchbowl Fly Out Day to Talk AI and Cybersecurity - House Committee on Homeland Security
This Republican knows Mythos. He's scared. - Punchbowl News
Anthropic's Mythos model raises alarm for Rep. Andrew Garbarino over AI threats to bank security - CryptoBriefing
'They showed us what it could do, and it's scary': Congressman on Anthropic Mythos model demo - WION News