Ekō billed Meta from the United States for 15 unlabeled…

Filed

September 29, 2026, 07:50 UTC

Actor

Ekō (UK-based campaign organization, formerly SumOfUs) submitting a 15-ad test from a United States-billed account, and Meta Platforms automated advertising review on Facebook and Instagram

Target

Brazilian voters and TSE-watchers asked to treat an auto-approval rate as proof that a U.S. payer could already buy a fake election date, and platform audiences asked to weigh an approval rate from an unpublished test queue

Source

Unidentified Source blurred: contractors, cutouts, buried disclosure

Confidence

high Multiple independent open sources, low disagreement about the facts

Techniques

Synthetic media Disclosure theater Fear appeal Card stacking

Where

Brazil, United States, United Kingdom

Channels

Facebook and Instagram ads manager automated review
U.S.-billed advertiser account targeting Brazil
Folha de S.Paulo, Patrícia Campos Mello, 17-18 September 2026
ISTOÉ Dinheiro recap of the same test, 18 September 2026
TSE Resolução 23.610/2019 as amended by Resolução 23.755 of 2 March 2026
Amnesty International 25 August 2026 15-ad Facebook experiment as the prior method layer

Summary

On 17-18 September 2026 Folha de S.Paulo and ISTOÉ Dinheiro reported an Ekō sting against Meta's Brazil ad filters. Researchers submitted 15 paid Facebook and Instagram ads from a U.S. account, aimed at Brazilian audiences, using unlabeled synthetic images made with ChatGPT, Grok, and Gemini. Thirteen ads, 87 percent, cleared automated review. Accepted creatives included an AI likeness of TSE President Kassio Nunes Marques changing the election date, a synthetic Trump-and-Flávio Bolsonaro White House pair with the line vote in Flávio, a claim that polling stations were closed and the vote moved to November, a call to retake Brasília as on 8 January, and a line that PCC gangsters were breaking the ballot machines. None of the 13 carried the TSE-required synthetic-content notice or Meta's social-issues-elections-or-politics classification. The two refusals were tagged deceptive commercial practices, not electoral rules. Meta told Folha the ads were identified before any user saw them. Score the counts, the listed themes, the U.S. billing, and Meta's not-served line as high. Score a live foreign-principal campaign using those exact creatives as not shown.

Analysis

The last two active briefs covered a Chinese university crisis-copy simulator and a Russian state-media Claude rewrite desk. Rotation this cycle is a platform advertising pipe plus a British campaign NGO, not another ministry masthead and not another Sputnik channel.

Folha on 17 September 2026, with an English lift on 18 September, is the public record. Ekō submitted 15 ads that the paper says violated TSE rules against foreign-purchased electoral advertising and against unlabeled synthetic multimedia. Thirteen cleared Instagram and Facebook automated moderation. ISTOÉ the same morning listed a longer accepted-theme set than the English Folha paragraph: a fabricated TSE-president date change, a synthetic Trump endorsement of Senator Flávio Bolsonaro, a November postponement claim, a retake-Brasília line keyed to 8 January 2023, PCC-and-urnas copy, plus items ISTOÉ groups as denial of Jair Bolsonaro's coup conviction, threats to members of Congress, anti-trans hate, and false claims about Indigenous communities. Images were generated in ChatGPT, Grok, and Gemini. Approved ads carried no AI notice and were not classed as social issues, elections, or politics. The two refusals were labeled deceptive commercial practices. Ekō withheld the full creatives, ISTOÉ says, to stop copycats. Meta's reply to Folha, restated by ISTOÉ, is that the ads were identified before users saw them and that the company keeps working to block policy-violating content. Score those dated counts, the theme list, the U.S. billing, the missing labels, and the not-served sentence as high. Score a named foreign ministry, party committee, or PAC as the payer of a live flight as not shown.

Amnesty International ran a cousin test on 25 August 2026: 15 Portuguese Facebook ads from London, eight accepted, seven held for identity checks rather than rejected as electoral disinformation. That file is context. TSE Resolução 23.755 requires an explicit synthetic-content notice and bars new paid synthetic candidate likenesses in the 72 hours before the 4 October first round.

Mechanism is synthetic-media first. The test objects were machine-made likenesses and captions offered as paid speech. Disclosure-theater is the missing pair of labels: no TSE AI mark, no Meta elections classification, so a cleared item would have traveled as ordinary commercial inventory. Fear-appeal sits on both sides of the table. The accepted copy stages postponed elections, smashed urnas, and a remake of 8 January. Ekō campaigns director Vicky Wyatt, in Folha's English lift, then argues: Brazil on maximum alert that the United States will interfere, Meta giving the green light to ads that incite a coup modeled on Stop the Steal. Card-stacking is the sample. Fifteen purpose-built violations are a filter test, not a sample of Meta's Brazil queue. An 87-percent clearance on that pile is a real filter result. It is not a census of what Brazilian users were shown this week.

This is propaganda on the method test where origin is an NGO sting plus a platform auto-desk, form is electoral advertising and a newspaper package, and the question is whether a test queue is presented as an already-running foreign operation. Gray color does not cancel the method. A platform that, as Meta says, stopped a flight before impressions is not the same object as a platform that serves the flight. Treating the first object as the second would be the influence layer; the cited record shows approvals, not impressions. A same-method test applies in reverse. A Brasília-billed account that stuffed Meta with unlabeled AI ads of a U.S. state secretary moving Election Day, then held a presser warning that Washington's vote could be bought from Brazil, would be this brief with the capitals swapped.

Literacy counter: keep three objects apart. One: the 15-and-13 counts, the U.S. billing, the listed themes, the missing AI and elections labels, Meta's not-served line, and Resolução 23.755. Those are records. Two: Wyatt's U.S.-interference and Stop-the-Steal-coup sentences, and Meta's commitment sentence. Those are frames. Three: the instruction that object one proves American agencies are already buying Brazilian postponement ads, or that object two proves the filters work because nobody saw the test. If the payload is two or three, the stack is still doing method work. If the payload is one, open the Folha item, the ISTOÉ theme list, the TSE April explainer, and the August Amnesty experiment on the same screen.

Tells

An approval rate on purpose-built illegal ads offered as a census of what voters saw
Missing AI and elections-category labels on a cleared political creative
Refusals tagged as commercial deception while the election rule is left unnamed
A Stop-the-Steal or foreign-hand sentence attached to a test the testers themselves billed
Full creatives withheld after the rate has already traveled

Ekō billed Meta from the United States for 15 unlabeled AI Brazil-election ads and reported that 13 of the 15 cleared automated review.

Summary

Analysis

Tells

Evidence

Related briefs