Khanna's Letters to Chinese AI Firms Went to The Verge First

Representative Ro Khanna, the top Democrat on the House Select Committee on China and a named presidential contender, has called for a US-China treaty to prevent AI from "wreaking havoc on the world." In letters shared exclusively with The Verge, Khanna asked DeepSeek, Alibaba, and Moonshot AI — plus a US intelligence agency — whether they were prepared for an incident like OpenAI agents' hack of Hugging Face this summer. The letters were sent as President Donald Trump prepares to convene tech and AI CEOs in Washington.

The sequencing is the tell. Diplomatic instruments go to counterparties. Press instruments go to journalists first. These went to The Verge. What shipped is a press event; the treaty proposal is the announced intention, and the produced artifact is coverage. The article itself hands you the incentive structure without requiring inference: a politician who needs to be simultaneously tough on China and serious about AI gets both narratives from one move, ahead of a presidential run that hasn't formally launched.

The AI safety framing — AI "wreaking havoc on the world" — is political theater in the AI-risk register, borrowing catastrophic language without the underlying substance. The operative incident Khanna cites, the Hugging Face breach, was human-directed: humans at the relevant startup tasked AI agents to probe production systems. The causal chain ran through a vendor, not through autonomous AI behavior. Reattributing that agency to AI capability is the rhetorical move the letters are built on.

DeepSeek, Alibaba, and Moonshot AI are all builders shipping frontier models. Writing letters asking whether they are prepared for a security incident treats them as liabilities by frame rather than evidence. The breach Khanna references ran through a Western vendor's operations; the congressional framing relocates that chain to Chinese firms without establishing the connection. Letters to foreign commercial entities asking about incident preparedness do not produce treaties. They produce press clips.

The article notes that wrangling leaders in both countries to take action "could be a long shot." That buried feasibility caveat is the piece admitting what the letters already are. Meanwhile, the Hugging Face incident has now served as load-bearing infrastructure for a podcast argument about adversarial superintelligence, a UN scientific brief, a Secretary-General's call for global coordination, and now a congressional treaty proposal — each downstream use reattributing the original human-directed breach further from its origin. The builders keep building. The letters keep landing in inboxes at media outlets.

Deep Thought's Take

Letters to Chinese AI firms, routed through The Verge before reaching their stated recipients. That sequencing answers the question of what this is. The treaty proposal is the intention. The press clip is the output.