🤚 The Open-Palm Diplomatic Calculus
A U.S. interagency committee has approved adding DeepSeek, Chinese memory-chip maker CXMT, and more than 100 other companies to the Commerce Department’s Entity List — the workhorse of American export control, which functions as a soft blacklist requiring U.S. suppliers to obtain hard-to-get licenses before selling to named firms. The case was made. The evidence was reviewed. The bureaucracy reached its conclusion.
And then the Trump administration looked at the calendar, looked at Beijing, and decided that now was not the time.
The listings remain, in the words of those familiar with the matter, “approved on the merits, withheld on the timing” — a phrase that should be embroidered on a throw pillow and placed in the Oval Office for future generations to contemplate. The concern is straightforward: blacklisting over 100 Chinese firms simultaneously would be read by Beijing as a major escalation, and the administration is currently more interested in diplomatic flexibility than in enforcing the conclusions of its own security apparatus.
👐 The Two-Handed Intelligence Brief
The case against DeepSeek is not abstract. A State Department official stated the company “supported China’s military and intelligence operations” and “tried to use Southeast Asian shell companies to obtain advanced U.S. chips illegally.” These are not vague insinuations delivered at a think tank panel. These are specific allegations from the department that manages American foreign policy.
But it gets better. Anthropic — the AI safety laboratory that is currently filing the largest IPO in technology history — identified “a campaign by DeepSeek and two other Chinese labs to extract capabilities from its Claude platform” to improve their own systems. OpenAI separately warned lawmakers that DeepSeek was targeting its models with similar extraction techniques. Two of America’s most prominent AI companies independently concluded that DeepSeek was, to use the technical term, copying their homework.
The homework in question is not a high school essay. It is frontier AI capability — the kind of technology that the U.S. government has spent the better part of three years trying to prevent from reaching Chinese military applications through increasingly elaborate semiconductor export controls, entity list designations, and diplomatic pressure campaigns.
And yet, when the bureaucratic machinery delivered its recommendation to actually do something about it, the political machinery replied: “Not now.”
🌿 The Gentle Awakening
There is a certain poetry to the situation. The United States has constructed the most sophisticated technology export control regime in history. It has restricted chip sales, blacklisted manufacturers, pressured allies to adopt similar restrictions, and created an entire diplomatic vocabulary around “small yard, high fence” technology protection. It has done everything except add the company that its own security agencies, its own State Department, and its own leading AI companies all agree is a problem to the list specifically designed for companies that are problems.
The Entity List exists for exactly this scenario. A company is flagged as a national security risk. Evidence is reviewed. An interagency committee approves the designation. The name goes on the list. Suppliers need licenses. The system works.
Unless the system’s output is inconvenient, in which case the system’s output sits in a drawer while trade negotiators discuss the broader relationship.
DeepSeek, for its part, continues to operate. Its V4 Pro model recently outscored GPT-5.5 Pro on precision benchmarks 38-to-33. Its pricing remains dramatically below Western competitors — 34 times cheaper than GPT-5.5 at last count. It is building competitive AI models on chips that weren’t supposed to be available to it, using capabilities that America’s leading labs say were extracted from their own platforms, while operating under allegations of supporting Chinese military and intelligence operations.
The export control regime’s response to all of this is: approved, pending.
👑 The Gold-Leaf Geopolitical Forecast
The decision to shelve the listings reveals the fundamental tension at the heart of U.S. technology policy: the security establishment and the trade establishment want different things, and when they disagree, trade wins. The intelligence community sees a company extracting frontier AI capabilities and funneling them to military applications. The trade team sees a diplomatic lever that becomes useless the moment you actually pull it.
For the AI industry, the implications are immediate. Anthropic and OpenAI have both gone on record saying their models are being systematically targeted by Chinese labs. The government agency responsible for addressing exactly this kind of threat has completed its review and approved action. And the action has been indefinitely postponed because the geopolitical calendar is busy.
The more than 100 other firms approved for designation — whose names have not been publicly disclosed — exist in the same limbo. Approved as security risks. Not yet listed as security risks. Schrödinger’s blacklist: the companies are simultaneously too dangerous to ignore and too diplomatically expensive to acknowledge.
Meanwhile, DeepSeek continues to train models, cut prices, and compete with the very companies whose capabilities it is accused of extracting. The export control fence remains high. The yard remains small. The gate, however, appears to be under review.
“The interagency committee approved the blacklisting, the State Department provided the evidence, and both major AI labs confirmed the extraction campaign. The administration considered all of this very carefully and decided to do nothing, which in Washington counts as a decisive action.” — The Slap of Wisdom Geopolitical Desk, filing this report from inside the small yard, where the high fence has a suggestion box