---
title: PRC-linked shops used Chinese open models and rented American voices so bill-shock copy would read as neighbor speech against U.S. data centers.
url: https://propagandaarchive.org/briefs/20260918-2010Z-prc-datacenter-agents/
description: On 18 September 2026 The New York Times, restated by Ynet, said unnamed U.S. officials described a China state-backed experiment pairing downloadable…
section: Watch briefs
---

![Illustration: an interwar radio-and-newsreel intelligence collage with a 1930s map of Europe and propaganda poster proofs.](https://propagandaarchive.org/static/generated/detail/hero-interwar.jpg)

[Watch briefs](https://propagandaarchive.org/briefs/) · September 18, 2026

# PRC-linked shops used Chinese open models and rented American voices so bill-shock copy would read as neighbor speech against U.S. data centers.

- **Filed:** September 18, 2026, 20:10 UTC
- **Actor:** Unnamed PRC-origin commercial influence cluster assessed by OpenAI as a private Chinese technology company serving provincial government clients, restated on 18 September 2026 by unnamed U.S. officials speaking to The New York Times as a state-backed autonomous-agent experiment using Chinese open-source models
- **Target:** U.S. platform users, local officials, and midterm audiences asked to treat American-looking bill-shock comments and cartoons as homegrown energy speech rather than as an undeclared foreign production line
- **Source:** Unidentified Source blurred: contractors, cutouts, buried disclosure
- **Confidence:** **high** Multiple independent open sources, low disagreement about the facts
- **Techniques:** [Astroturfing](https://propagandaarchive.org/techniques/astroturf/) [Synthetic media](https://propagandaarchive.org/techniques/synthetic-media/) [Card stacking](https://propagandaarchive.org/techniques/card-stacking/) [Bandwagon](https://propagandaarchive.org/techniques/bandwagon/) [Narrative laundering](https://propagandaarchive.org/techniques/narrative-laundering/)
- **Where:** China, United States
- **Channels:** ChatGPT accounts accessed by VPN and prompted in Simplified ChineseInstagram, Facebook, X, and TikTok inauthentic accounts and agentsOpenAI June 2026 threat report, Data Center Bandwagon and Tech and Tariffs clustersX Global Government Affairs disclosure, 28 August 2026New York Times, Sheera Frenkel, Eli Tan, and Dustin Volz, 18 September 2026Ynet restatement of the same Times file, 18 September 2026House Foreign Affairs Subcommittee testimony by Sam Lyman, 15 September 2026

## Summary

On 18 September 2026 The New York Times, restated by Ynet, said unnamed U.S. officials described a China state-backed experiment pairing downloadable Chinese open-source models with autonomous agents to open and post from fake accounts on Instagram, Facebook, X, and TikTok. That method sits on two earlier inventories. OpenAI on 10 June 2026 banned ChatGPT clusters labeled Data Center Bandwagon and Tech and Tariffs, assessed as likely China-origin, run by a private tech firm serving provincial clients; operators prompted in Simplified Chinese through VPNs, posed as Americans, and generated English comments and comics on power bills. X Global Government Affairs on 28 August 2026 said Safety had found a farm of ~200,000 suspected Chinese inauthentic accounts, ~200 of which posted that data centers raise electricity prices and strain the grid, plus AI cartoons of operators enriching themselves. Clemson researcher Darren Linvill told Business Insider that sampled posts had little or no engagement. Score the OpenAI ban, the X counts, and the Times/Ynet method description as high. Score a named MSS or CAC tasking order, and any proof the farm moved a permit or a midterm, as not shown.

## Analysis

The last two active briefs covered an Iran-attributed autonomous-agent farm on U.S. platforms and an SVR election-day mirror brief. Rotation this cycle is the China slice of the same 18 September method file, not another ministry press bureau and not another Huanyu rehearsal model.

Three records have to stay separate. OpenAI on 10 June 2026 is the primary on production: two ChatGPT clusters likely originating in China; Data Center Bandwagon generated English comments and comics claiming AI data centers raise family electricity bills; Tech and Tariffs generated cartoons on tariffs and U.S. tech dominance and told the model to keep Xi Jinping out of the frame. OpenAI assessed the operators as a social-media team at a private Chinese technology company working for provincial-level government clients, accessing the product by VPN and prompting in Simplified Chinese while posing as Americans. The same report said the operation tried to ride an existing debate and found no evidence of meaningful breakout beyond its own activity. X Global Government Affairs on 28 August 2026 is the primary on inventory: about 200,000 suspected Chinese inauthentic accounts; about 200 inside that farm posting that data centers drive household prices and strain the grid, plus AI-generated cartoons of operators with cash bags. Darren Linvill, using Clemson holdings of suspended accounts, told Business Insider the same day that sampled items had zero engagement and that many accounts looked deactivated before they built a following. The New York Times on 18 September 2026, restated by Ynet, is the primary on the newer agent layer: unnamed U.S. officials said China, like Iran, had begun pairing Chinese open-source models with autonomous agents that could open accounts and coordinate posts with safeguards against deception turned off. Ynet said fewer details were offered on the Chinese experiment than on the Iran farm. Score those dated bans, counts, and the official-sourced agent description as high. Score a named ministry letterhead, a public model card for the agents, and any causal claim that the stack stopped a Utah or Louisiana permit or moved a November race as not shown.

White-channel state media is a related object, not this actor. CGTN in October 2025, China Daily on 12 March 2026, and Global Times in April 2026 ran open English copy on U.S. data-center power costs. That is disclosed Party-state commentary. Do not collapse it with the VPN-prompted American costume.

Mechanism is astroturf first. Accounts that claim to be ratepayers in the Mid-Atlantic are rented grassroots. Synthetic-media is the production line: models write the comics; agents, on the Times account, operate the software. Card-stacking is the input list: Bloomberg-adjacent bill spikes and capacity-auction clips in; China's own data-center build, including official reporting on Ulanqab-scale load, and OpenAI's no-breakout sentence out. Bandwagon is the campaign's own name and the warrant: treat rising comment volume as proof the neighborhood already decided. Narrative-laundering is the commercial cutout. A provincial client does not have to sign the cartoon if a private shop and an American bio do the posting.

This is propaganda on the method test where origin is an undeclared PRC-origin commercial cluster, form is vernacular U.S. energy speech, and the design goal is to move identification so a foreign desk reads as a neighbor with a power bill. Gray color does not cancel the method. Real Americans arguing that a county should reject a substation is not the violation. Standing up American-looking accounts, then letting software post bill-shock comics as if those people lived next to the fence, is the influence layer. A same-method test applies in reverse. A Virginia contractor that used an open-source model and a fleet of agents to pose as Inner Mongolia villagers, tagged county officials in Ulanqab, and sold the resulting comment stack as domestic dissent would be this brief with the substations swapped.

Literacy counter: keep three objects apart. One: the 10 June OpenAI clusters and no-breakout line, the 28 August X 200,000-and-200 counts, Linvill's engagement check, and the 18 September Times/Ynet agent-and-model description. Those are records. Two: the midterm-threat sentence and the first-of-their-kind label. Those are frames. Three: the instruction that object one proves every local bill complaint is MSS product, or that object two proves the farm never existed because the sampled bots had no likes. If the payload is two or three, the stack is still doing method work. If the payload is one, open the OpenAI June report, the X GlobalAffairs post, the Business Insider Linvill item, and the Times and Ynet 18 September pieces on the same screen.

## Tells

- Simplified-Chinese prompts and VPN access paired with American ratepayer bios
- Comics and short comments that cite real bill or auction stories while hiding the poster
- Safeguards against deceptive-account creation described as turned off
- A 200,000-account farm headline that travels faster than the 200-account theme slice or the engagement audit
- Open state-media bill copy offered later as independent confirmation of the covert costume

## Literacy counter

Keep three objects apart. One: the 10 June OpenAI clusters and no-breakout line, the 28 August X 200,000-and-200 counts, Linvill's engagement check, and the 18 September Times/Ynet agent-and-model description. Those are records. Two: the midterm-threat sentence and the first-of-their-kind label. Those are frames. Three: the instruction that object one proves every local bill complaint is MSS product, or that object two proves the farm never existed because the sampled bots had no likes. If the payload is two or three, the stack is still doing method work. If the payload is one, open the OpenAI June report, the X GlobalAffairs post, the Business Insider Linvill item, and the Times and Ynet 18 September pieces on the same screen.

## Evidence

1. [New York Times, 18 Sep 2026: Frenkel, Tan, Volz on Iran, China, and Israeli-firm autonomous-agent campaigns](https://www.nytimes.com/2026/09/18/technology/iran-china-autonomous-ai-influence-campaigns.html)
2. [Ynet, 18 Sep 2026: restates Times file; fewer details on the Chinese experiment than on the Iran farm](https://www.ynetnews.com/tech-and-digital/article/sy9x2aqfml)
3. [OpenAI, 10 Jun 2026: PRC-linked Data Center Bandwagon and Tech and Tariffs clusters; no meaningful breakout](https://openai.com/index/prc-linked-influence-operations-ai-debates/)
4. [OpenAI June 2026 threat-report PDF hosting the same two clusters](https://cdn.openai.com/pdf/96b559fa-c165-4575-805d-e636909e2f78/June-2026-Threat-Report.pdf)
5. [X Global Government Affairs, 28 Aug 2026: ~200,000 suspected Chinese inauthentic accounts; ~200 on data-center copy](https://x.com/GlobalAffairs/status/2093130747796148634)
6. [Business Insider, 28 Aug 2026: Linvill on sampled X bots with no engagement or followers](https://www.businessinsider.com/x-chinese-bots-data-center-debate-no-engagement-followers-2026-8)
7. [Axios, 10 Jun 2026: Nimmo line that the operation tried to interfere in an existing debate](https://www.axios.com/2026/06/10/openai-china-ai-data-center-tariffs-chatgpt)
8. [China Daily, 12 Mar 2026: open English state-media copy on U.S. data-center electricity bills](https://www.chinadaily.com.cn/a/202603/12/WS69b1f569a310d6866eb3d4f5.html)

## Related briefs

- Sep 18, 2026Unidentified — [Iran-linked agents used Chinese open-source models to stand up American-looking accounts so rented personas would read as U.S. political speech.](https://propagandaarchive.org/briefs/20260918-1810Z-iran-agent-farm/) — [Astroturfing](https://propagandaarchive.org/techniques/astroturf/) [Synthetic media](https://propagandaarchive.org/techniques/synthetic-media/) [Manufactured consensus](https://propagandaarchive.org/techniques/manufactured-consensus/) [Card stacking](https://propagandaarchive.org/techniques/card-stacking/) [Disclosure theater](https://propagandaarchive.org/techniques/disclosure-theater/)
- Sep 18, 2026Identified — [Sun Yat-sen University launched Huanyu Zhiying 1.0 so simulated foreign desks and silicon audiences would pre-tune China's crisis copy.](https://propagandaarchive.org/briefs/20260918-1440Z-huanyu-zhiying/) — [Synthetic media](https://propagandaarchive.org/techniques/synthetic-media/) [Manufactured consensus](https://propagandaarchive.org/techniques/manufactured-consensus/) [Narrative laundering](https://propagandaarchive.org/techniques/narrative-laundering/) [Euphemism / doublespeak](https://propagandaarchive.org/techniques/euphemism/) [Card stacking](https://propagandaarchive.org/techniques/card-stacking/)
- Sep 18, 2026Identified — [A House subcommittee titled its hearing Trojan Horses so China-Iran-Cuba influence would read as a hidden army inside U.S. nonprofits and streamers.](https://propagandaarchive.org/briefs/20260918-0208Z-house-trojan-horses/) — [Semantic hijacking](https://propagandaarchive.org/techniques/semantic-hijack/) [Name-calling](https://propagandaarchive.org/techniques/name-calling/) [Card stacking](https://propagandaarchive.org/techniques/card-stacking/) [Testimonial](https://propagandaarchive.org/techniques/testimonial/) [Transfer](https://propagandaarchive.org/techniques/transfer/)