A viral video of Washington County chair candidate Nafisa Fai in front of an upside-down U.S. flag spread nationwide. A local outlet called it a deepfake, then unpublished its story a day later after new images showed the flag really had been upside down — a case study in how hard video is to verify in real time.
2026-10-20 · by Moe Ameen
A roughly 26-second clip of Nafisa Fai — a Washington County, Oregon commissioner running for chair of the county Board of Commissioners — went viral the week of October 19, 2026 and turned into a textbook example of how hard it is to establish what is real when a short clip becomes the subject of dueling authenticity claims faster than anyone can check them. The footage came from an October 10 community event in Minnesota where Fai spoke in Somali; a selectively edited version, showing an upside-down American flag behind her, was posted by the right-wing X account "End Wokeness" (roughly 3.9 million followers) and amplified by other prominent accounts. The upside-down flag — a distress and protest symbol — was used to frame her as un-American, and the post drew vitriolic responses, including calls for her deportation and a confrontation after a candidates forum in Hillsboro.
The Oregon Journalism Project published a story describing the clip as a "deepfake" circulated to smear Fai, citing screenshots her campaign provided that showed the flag right-side up and a witness who said the flag at the event had been displayed upright and that the viral clip had been edited. Willamette Week and Lookout Eugene-Springfield republished the piece. Within about a day the reporting fell apart: after readers flagged the conflicting images, the Oregon Journalism Project obtained new images showing the flag "had in fact been upside down" at the event, unpublished the original story, and issued an editor's note on October 20 saying it "regrets the error." The outlets that had run it also apologized.
The dispute did not end there. The Oregon GOP and conservative outlets argued the opposite of the original story — that the upright-flag image Fai's team supplied was itself AI-generated, pointing to an alleged SynthID watermark they said identified the image as originating from an OpenAI tool. That claim circulated on partisan accounts and has not been independently verified here; OpenAI's own guidance on these signals says a missing watermark does not prove an image wasn't made with its tools — metadata can be stripped by editing or format conversion and the watermark itself can degrade — so a detected signal points to an OpenAI tool, but a missing one proves nothing either way. The net result is that, well into a live campaign, there was no quick, authoritative answer to a basic question — which version of a short clip reflected reality — and a professional newsroom publicly got it wrong before correcting course. Fai is on the November 3, 2026 ballot for chair of a board that oversees a roughly $2.1 billion county budget.
The lesson of this episode is not "disclose your AI" — it is that in a world where a short clip can be edited to mean the opposite of reality and even a newsroom cannot verify it in time, the durable defense is owning a loud, consistent, unmistakably-yours record of what you actually say. A [deepfake](/glossary/deepfake) or a doctored clip works by filling a vacuum; a channel that posts your real positions, in your real voice, every week leaves no vacuum to fill — and gives your audience an authoritative place to check a viral claim against. That record is a content-production problem, and it is exactly what [Kompozy](/) is built to produce at volume.
Point Kompozy at footage you control — your event recording, your talk, your statement to camera — and it generates net-new, first-party content from an identity you own rather than a scraped likeness of someone else: a face-locked [Persona Short](/glossary/persona-shorts) in your own voice, carousels and quote graphics pulled from your actual words, a blog post, and an email newsletter, all held to one tone and look by a single [Persona Brief](/glossary/persona-brief) so every piece reads as verifiably you. [Autopilot](/glossary/autopilot) then schedules and publishes the batch across the eight social platforms plus blog and email from one queue, each asset behind a per-post [review gate](/glossary/quality-gates) you clear before it ships. The image models Kompozy uses (OpenAI's gpt-image) embed [C2PA](/glossary/c2pa) and SynthID provenance signals on what you generate, so your own assets carry the authenticity markers the platforms increasingly read. None of that stops someone from editing a clip of you — nothing does. What it does is make sure that when a fake spreads, the real you is already everywhere, consistent, and easy to find.
A roughly 26-second clip of Washington County, Oregon chair candidate Nafisa Fai speaking at an October 10, 2026 event in Minnesota — showing an upside-down American flag behind her — went viral via a large right-wing X account. The Oregon Journalism Project published a story calling it a "deepfake," then unpublished it on October 20 after obtaining new images showing the flag had in fact been upside down, saying it "regrets the error."
That remained contested, and the newsroom that first called it a deepfake retracted that claim after new images showed the flag genuinely had been upside down at the event — meaning the viral clip appears to have been authentic rather than a fabrication. Partisan accounts separately claimed the campaign's own rebuttal image was AI-generated, citing an alleged watermark that was not independently verified. The episode's point is that no one could quickly establish the truth.
Oregon is among the roughly 30 US states that have passed laws addressing AI or deepfakes in election content, most of which require disclosure; a few states — including Maryland, Minnesota, and Texas — go further and ban certain deceptive deepfakes outright. There is no federal law specifically governing AI-generated political media, so requirements vary by state. Whether a given disputed clip qualifies as regulated campaign communication is itself often unclear.
You cannot stop someone from editing or fabricating a clip, but you can make the real you easy to verify: keep a loud, consistent, first-party record of what you actually say across your own channels, generate content from an identity you own rather than a borrowed likeness, and favor assets that carry provenance signals like C2PA or SynthID. A tool like Kompozy produces that first-party record at volume and publishes it across platforms from one queue.