I am writing as an alarmed constituent about a national security threat that demands immediate congressional action.
On April 7, 2026, Anthropic announced Project Glasswing — built around a stunning admission: they have created an AI model so dangerous they cannot release it to the public. That model, Claude Mythos Preview, autonomously finds and exploits critical vulnerabilities in every major operating system and browser — flaws human experts missed for decades — at a cost of just $50 in computing power.
Anthropic has already privately briefed senior government officials that this technology is expected to significantly increase the likelihood of large-scale cyberattacks in 2026. The systems at risk run our power grids, hospitals, and financial infrastructure.
What alarms me most is this: the private sector is racing ahead while Congress has not acted. CISA's incident-reporting rule remains unfinished. There is no independent oversight body. And fewer than 1% of the vulnerabilities this AI has already found have been patched.
I urge you to:
1. Convene immediate hearings on AI-driven cybersecurity risks.
2. Mandate independent verification of safety claims from Anthropic and its partners.
3. Fund and finalize the DHS AI Information Sharing and Analysis Center and the National Cyber Director's AI security framework.
4. Establish mandatory oversight for any AI capable of autonomous cyberattack development.
Anthropic itself has said this problem requires government involvement. That moment is now — not after a catastrophic attack.
Please act with the urgency this threat demands.
Sign Petition
