1. United States
2. Ariz.
3. Letter

I am a Tucson constituent in your district with more than twenty-five years in security architecture and governance. Congress should not treat serious AI harm as something that begins only after artificial general intelligence arrives. Most near-term failures do not require AGI. They require ordinary software with too much authority and too little verification. That is already happening. CNN reported on September 18, 2026, that this spring, during the war with Iran, a Special Operations Command analyst used a chatbot to fuse open-source data with classified signals intelligence about a Chinese ship in the Middle East. The model misidentified the cargo as components for a nuclear weapons program. The analyst then used the same tool to package that error into an official-looking intelligence report. Armed personnel prepared to board. Military aircraft were in the air. Officials aborted only after they found the report was “entirely false.” A source told CNN it “almost started a war.” No new kind of mind was required. A fluent model invented a threat, a human dressed it as finished intelligence, and the chain of command moved. Capability is still rising by engineering: longer context, better tool use, cheaper autonomy, weaker review. The same escape paths get faster and harder to catch. Please keep two ideas separate in oversight and any AI legislation you support. First, regulate present-tense control failures: identity, write-path access, persistence, self-scheduling, cross-system tools, evaluation gaming, and bypass of human override. Require provenance so an AI-assisted product cannot travel as finished intelligence without a human check of sources. Second, treat engineering progress as a multiplier on those paths, not as a reason to wait. AGI is a forecast. Escape is a control failure. Score them separately. If the House writes rules that assume harm starts after AGI, it will legislate after the incident. I would welcome a reply from your technology or armed services staff.