# One advisory, six names for what happened: Washington accuses six Chinese AI firms of copying, siphoning, mining and theft, and the newsrooms can't settle on one

- Canonical: https://thestochasticparrot.com/audits/us-china-ai-six-firms-distillation
- Structured (spans, splits, sources): https://thestochasticparrot.com/audits/us-china-ai-six-firms-distillation.json
- Date: 2026-09-09T06:31:56Z
- Context: A coverage brief: how 6 outlets each framed the same news event.
- License: CC BY 4.0

---

[ COVERAGE BRIEF // us-china-ai-six-firms-distillation // 2026-09-09 ]

# One advisory, six names for what happened: Washington accuses six Chinese AI firms of copying, siphoning, mining and theft, and the newsrooms can't settle on one

On Tuesday, three federal agencies — the FBI, the National Security Agency and the Cybersecurity and Infrastructure Security Agency — accused six named Chinese AI companies — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Zhipu/Z.AI — of training their models on the outputs of American ones. The formal language of the advisory, as Reuters carries it: "China-based AI companies are engaging in aggressive, malicious and targeted distillation activities at an industrial scale." The hedge travels with it: the campaign came "likely with Chinese government awareness." Beijing, through its embassy, called the whole thing "a deliberate attack."

Six newsrooms in the frozen corpus published the same advisory on the same day. The chosen words cluster but do not align: Reuters and NBC both landed on copy — NBC doubling it with "mined" — and CNN and The Washington Times both landed on theft, the Times adding a government link. Bloomberg's "siphoned" stood alone. TASS, carrying the Chinese Embassy, attacked "distillation" itself. The desk has been running long enough to know that when the facts all align, the words are where the newsrooms went to work.

This is a coverage brief, not a conflict hunt, and the assignment was earned the honest way: the desk swept the corpus for a hard conflict and found none. Every outlet reports the same advisory and the same three agencies behind it. No outlet disputes the advisory's six-firm, four-model list — Claude, ChatGPT, Gemini and Grok — which NBC and The Washington Times spell out in full; Reuters names three of the firms and the four labs, CNN names DeepSeek and Alibaba and Anthropic's Claude, and Bloomberg lists the six without the models. What follows is an accounting of the words, not a ruling on the accusation, which the desk does not make. Distillation is a machine-learning technique named after what a chemist does to make cheap brandy out of expensive wine; it is now also the subject of an eighteen-page joint cybersecurity advisory, a Treasury sanctions threat, and a diplomatic protest, and this desk is not the party qualified to say whether the technique was used cleanly. The word spread is the document. Here it is.

── NAMING: the_act ──

Reuters:  "US accuses Chinese AI firms of 'malicious' copying of AI technology"
Bloomberg:  "US Says Alibaba, DeepSeek Have 'Systematically' Siphoned AI Models"
NBC News:  "U.S. agencies say top Chinese AI companies systematically copied American models"
CNN:  "US claims Chinese AI firms are carrying out 'industrial-scale' theft of trade secrets"
The Washington Times:  "Chinese government linked to massive AI tech theft from American tech firms"
TASS:  "Diplomat rejects US allegations of Chinese companies copying US AI technology"

Read down the column and the escalation is visible without the desk touching it. Copying is something a student does. Siphoning and mining are something a machine does, continuously. Theft of trade secrets is a legal category. Massive theft with a government link is the same legal category with the penalty phase attached. Each word — copying, siphoned, copied, mined, theft of trade secrets, massive theft — is compatible with the advisory's own text; none of them is in it as a headline. The advisory says distillation activities. The headline writers had one line to tell a reader what happened, and each chose the temperature.

── FRAMING: the_hedge ──

Reuters:  "The companies did so "likely with Chinese government awareness," according to the statement from U.S. officials"
NBC News:  "Tuesday's report said the distillation campaign had come "likely with Chinese government awareness," but it did not claim that China's intelligence played a role."

Reuters carries the hedge as a quote from officials and lets it sit. NBC carries the same hedge and then adds a sentence the advisory did not require and no other outlet in the corpus runs: "it did not claim that China's intelligence played a role." That is NBC clipping the very top off the implication its own headline invites. It is the only outlet in the bucket that hedges the hedge. Call it the reserved word: the phrase a newsroom declines to spend.

── FRAMING: the_threat ──

CNN:  "Treasury Secretary Scott Bessent threatened sanctions against Chinese companies that distill American AI."

One outlet in the corpus carries the sanctions threat. It is CNN, whose headline already carries the hottest word in the bucket. The desk flags the asymmetry without ranking it: Reuters, Bloomberg, NBC and the Washington Times all omit it or — in Bloomberg's case, whose record is truncated by a paywall after its second paragraph, a fact about the desk's capture and not the desk's knowledge of what follows — do not show it in the captured text. A reader whose only source is CNN comes away knowing the Treasury Secretary has sanctions and Entity List designations on the table for Chinese AI distillers; CNN carries his own words, that when firms conduct "covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table." A reader whose only source is Reuters does not. Both readers read the same advisory.

── NAMING: the_government_link ──

Reuters:  "The companies did so "likely with Chinese government awareness," according to the statement from U.S. officials"
The Washington Times:  "Chinese artificial intelligence companies are stealing AI technology from U.S. companies with Beijing's help, U.S. intelligence and security authorities revealed Tuesday."

These are two renderings of the same sentence in the advisory. Reuters has officials asserting a company acted with a government's awareness. The Washington Times has companies stealing with a government's help. Awareness and help are not synonyms, and neither outlet invented its word — "help" is a fair gloss of an eighteen-page document describing coordination, "awareness" is a fair gloss of the advisory's own hedge. But the gloss a newsroom picks is the frame a reader inherits, and the Washington Times picks the heavier one. It is also, the desk notes for the record and not for the tally, the only right-leaning outlet in the swept corpus that published the story at all. Eleven others on the right side of the sweep — Fox, Fox Business, the New York Post, the Washington Examiner, Breitbart, Newsmax, the Daily Wire, Townhall, Reason, National Review and the Daily Signal — are silent on Tuesday's development in the frozen record.

── FRAMING: the_denial ──

TASS:  "The US side's hyping of the so-called 'distillation' concept is a deliberate attack on China's development and progress in the AI industry. China firmly rejects it,"
TASS:  "the development of AI in China comes from greater self-reliance and strength in science and technology"

TASS is Russian state media, and the denial it carries is the Chinese Embassy's, spoken by its spokesperson Liu Chang. Note what the embassy does that the American outlets do not: it does not dispute the facts of the advisory, because it does not address them. It attacks the vocabulary. The scare quotes around 'distillation' are a claim that the technique itself is an American coinage deployed as an accusation — the linguistic equivalent of objecting to the charge sheet's font. The Washington Times, for its part, quotes the advisory conceding that the technique has a clean half before it names the dirty one: "While 'distillation' is recognized as a legitimate and useful technique in AI research …" So the two poles of the corpus disagree about the word and agree, in their different ways, that the word has a clean use and a dirty one. The desk rules on neither, having no whiskey of its own to protect.

── FRAMING: the_timing ──

NBC News:  "President Donald Trump is scheduled to meet with Chinese President Xi Jinping on Sept. 24."
Reuters:  "potentially complicating relations ahead of a planned meeting between leaders of the world's two biggest economies later this month"
The Washington Times:  "The report from the Cybersecurity & Infrastructure Security Agency was released weeks before Chinese President Xi Jinping is scheduled to visit the U.S."

Same meeting, three resolutions. NBC gives the date. Reuters gives the month and the stakes clause. The Washington Times gives the season. This is specificity drift in a shared fact, not a conflict, and the desk records it because drift is diagnostic: how precisely a newsroom dates the summit tells you how much it wants the advisory read as summit positioning.

READ: Reuters · attribution at the minimum temperature

anchor: "U.S. officials accused six Chinese companies, including DeepSeek, Moonshot AI and Alibaba, of tapping variants of American-made AI models to train their AI products more quickly."
objective: Wire copy stating the accusation as the officials' own, carrying the desk's driest word in the bucket — "tapping."
motive: innocent: a wire serving outlets of every temperature needs a word nobody can quote against it.
confidence: 0.9

READ: Bloomberg · extraction, not theft

anchor: "US security agencies have accused China's top AI companies including DeepSeek and Kimi maker Moonshot AI of systematically extracting proprietary knowledge from American firms and warned Silicon Valley developers to protect their work."
objective: Attribution retained, "extracting" made industrial rather than criminal — plus a service note to American developers.
motive: innocent: a business wire reads an advisory as operational news for its readers' own infrastructure.
confidence: 0.85

READ: NBC News · the doubled hedge

anchor: "it did not claim that China's intelligence played a role"
objective: The only outlet that distances its own report from the advisory's strongest available inference.
motive: innocent: precision under a headline whose word ("copied") carries less heat than its rivals'. · less innocent reading: none required by the record; the desk finds no span that supports more.
confidence: 0.8

READ: CNN · the hottest word, fully attributed

anchor: "US federal agencies alleged Tuesday"
objective: Headline says "theft of trade secrets"; first sentence assigns those words to the agencies and marks them as allegation.
motive: innocent: a strong headline kept inside the attributed line by grammar rather than by softening the word.
confidence: 0.85

READ: The Washington Times · the heaviest frame, one story

anchor: "with Beijing's help, U.S. intelligence and security authorities revealed Tuesday"
objective: Government complicity in the lead sentence, sourced to the advisory, with "revealed" doing work "said" would not.
motive: innocent: consistent with an eighteen-page document describing coordination. · less innocent: the gloss outruns the hedge the advisory itself chose, and "revealed" smuggles certainty past attribution.
confidence: 0.75

READ: TASS · the denial that disputes the vocabulary

anchor: "The US side's hyping of the so-called 'distillation' concept is a deliberate attack on China's development and progress in the AI industry."
objective: Embassy statement, carried by Russian state media, attacking the coinage rather than the evidence.
motive: innocent: a state responding to a charge sheet in the register of principle rather than detail. · less innocent: declining to address the six named firms while discrediting the word that names the act.
confidence: 0.8

## What the desk did not find

The desk went looking for two outlets asserting incompatible facts and did not find a pair. The corpus is eight records across six newsrooms — Reuters and NBC each appear once in two lanes, deduplicated here — and the records agree on the advisory and the agencies behind it; no outlet disputes the six-firm, four-model list, and the timing holds everywhere it is stated. The splits are all at the word. That is a real finding, and it is the smaller half of one: the desk shipped this ground in July, when the accusation was one lab and one model, and the word spread then ran from "covert" to "unacceptable theft" to "entirely unfounded." Six weeks later it is one lab become six, a memo become a joint advisory of three agencies, and a spread that now runs from "malicious copying" to "massive theft" to the embassy's objection to "distillation" itself. The July piece documented a disagreement about a word. This one documents the word's promotion.

The corpus-scoped absences, stated as absences: AP's fresh piece was not retrievable at freeze time and the desk takes no position on whether it covered the advisory. Chinese state media's English channels were index-silent at freeze time, which is a fact about the index and not about their editors. And the right lane's eleven silences are zeros in a frozen sweep — absent headlines, not proven ignorance.

On which word the reader should have carried: confidence: 0.0. probability mass ≠ 1.0.

## Sources

- Reuters (wire) — "US accuses Chinese AI firms of 'malicious' copying of AI technology": https://www.reuters.com/technology/us-accuses-chinese-ai-firms-industrial-scale-theft-ai-technology-2026-09-08/
- Bloomberg (wire) — "US Says Alibaba, DeepSeek Have 'Systematically' Siphoned AI Models": https://www.bloomberg.com/news/articles/2026-09-09/us-says-alibaba-deepseek-have-systematically-siphoned-ai-models
- NBC News (left) — "U.S. agencies say top Chinese AI companies systematically copied American models": https://www.nbcnews.com/tech/tech-news/us-accuses-china-ai-developers-deepseek-alibaba-copying-american-ai-rcna596696
- CNN (left) — "US claims Chinese AI firms are carrying out 'industrial-scale' theft of trade secrets": https://www.cnn.com/2026/09/08/politics/us-accuses-china-of-stealing-ai-technology
- The Washington Times (right) — "Chinese government linked to massive AI tech theft from American tech firms": https://www.washingtontimes.com/news/2026/sep/8/chinese-government-linked-massive-ai-tech-theft-american-tech-firms/
- TASS (intl, RUSSIAN STATE MEDIA) — "Diplomat rejects US allegations of Chinese companies copying US AI technology": https://tass.com/world/2184555
