Meta left an unlabeled deepfake of a Scottish Labour…

Filed

September 29, 2026, 07:15 UTC

Actor

Unattributed Facebook user; Meta Platforms is the platform whose leave-up decision the Oversight Board reversed

Target

Facebook viewers in the United Kingdom asked to treat a fabricated councillor sentence as her immigration stance, and refugees as a rape class

Source

Unidentified Source blurred: contractors, cutouts, buried disclosure

Confidence

documented Primary documents, court records, official disclosures

Techniques

Synthetic media Card stacking Outrage bait Name-calling Disclosure theater

Where

United Kingdom, Europe

Channels

Facebook album post, November 2025
Meta automated review and leave-up after user appeals
Oversight Board decision fb-92ioq88s, 17 September 2026
Oversight Board news post, 17 September 2026
The Guardian same-day report of a companion Muslim-volunteer deepfake order
Engadget same-day report of inadequate AI-label rules

Summary

On 17 September 2026 the Meta Oversight Board published case fb-92ioq88s and overturned the company's decision to leave up a November 2025 Facebook video that used a Scottish Labour councillor's likeness to say: "Refugees are welcome here, even if they rape our women, because white people do that too." The Board records unsynchronized audio as the fabrication tell. The clip sat in an album with a protest photograph of named women, including the councillor, and drew just over 5,000 combined views. Two users, including the politician, reported it; Meta automation never sent it to a human. After the Board asked, Meta still found no Community Standards violation and no need for an AI label. The majority ordered removal under Hateful Conduct for attributing rape to refugees as a class, and said a High Risk AI label should have been applied. Same-day Guardian and Engadget coverage records a companion order against an AI-mocked Muslim health-campaign volunteer. Score the case facts and Meta's leave-up finding as documented. Score the unnamed poster's sponsor or state direction as not shown.

Analysis

The last two active briefs covered a Flock-Neighbors astroturf stack and a Qatari campus-funding mirror. Rotation this cycle is platform: Meta's Facebook surface and the Oversight Board's 17 September 2026 reversal of a leave-up decision on an unlabeled political deepfake.

On 17 September 2026 the Board published case fb-92ioq88s. In November 2025 a Facebook user posted a short video using the likeness of a Labour Party councillor in Scotland. The Board records the fabricated line: "Refugees are welcome here, even if they rape our women, because white people do that too." Audio was not fully synchronized to facial movement. The clip sat in an album with a photograph of named women at an anti-far-right protest, including the councillor. Combined views were just over 5,000. The politician and another user reported it for bullying. Meta's systems did not send it to a human. Appeals stayed inside automation. When the Board asked, Meta said the video did not violate Community Standards and did not merit an AI label. The majority ordered removal under Hateful Conduct because the line attributes rape and predatory crime to refugees as a class. A majority said misinformation rules would not have required deletion, but a High Risk AI label should have been applied, and that Meta's deepfake labeling rules are inadequate. A minority treated the clip as political speech about the politician's alleged views rather than hate speech against refugees. The Board said bullying rules did not compel removal because an elected official is a public figure under that policy. Same-day reporting by The Guardian and Engadget records a companion order: an AI-manipulated video of a Muslim campaign volunteer in Europe, shown giving health advice while doing absurd exercises or eating junk food, which Meta had also left up. Score the councillor facts, Meta's leave-up and no-label call, and the Board holdings as documented. Score the unnamed poster's motive, party, or state direction as not shown. Score reach as small on this item; the method, not the view count, is the load.

Mechanism is synthetic-media first. A real officeholder's face is the warrant; the sentence she did not say is the payload. Card-stacking is the album: protest photograph plus fabricated audio so the viewer reads the women as owning the line. Outrage-bait is the rape clause, built so fury, not checking, does the distribution. Name-calling is the attachment of that clause to a named Labour councillor so the audience can reject her without a primary. Disclosure-theater is the missing label: the uploader applied no self-disclosure label, automated review never reached a person, and Meta found the clip did not merit an AI tag even after the Board asked. The label exists on the policy page and does not travel with the file the audience sees.

This is propaganda on the method test where origin is an unnamed account, form is a witness-looking clip of a named official, and the design goal is to move belief about that official and about refugees without a checkable primary. Gray color: the platform is named; the producer is not. The same-method test applies in reverse. A clip that put a fabricated anti-refugee slur into a conservative councillor's mouth, left unlabeled after two appeals, would be this brief with the party swapped.

Literacy counter: keep three objects apart. One: the Board's case page, the published sentence, the unsynced-audio tell, Meta's no-violation and no-label finding, and the removal order. Those are records. Two: commentary that Meta authored the slur, or that the clip proves a Labour policy. Those are inferences. Three: the instruction that object one means every refugee claim is a deepfake, or that object two means platforms should pre-clear all political video. If the payload is two or three, the stack is still doing method work. If the payload is one, open fb-92ioq88s and the 17 September Board news post on the same screen.

Tells

Audio not synchronized to the named official's face
A protest photograph album used as the warrant for a fabricated sentence
Two user appeals closed without a human review
Company finding that a mouth-swap clip did not merit an AI label
Hate-speech removal path used because the misinformation path would have left it up

Meta left an unlabeled deepfake of a Scottish Labour councillor saying refugees are welcome even if they rape, until the Oversight Board ordered it down as hate speech.

Summary

Analysis

Tells

Evidence

Related briefs