AI Watermark in Washington County Candidate Image Draws Into Question Reporting Behind Oregon Democrats’ “Racist Deepfake” Claims

Oregon — Oregon Democrats and Democratic elected officials amplified an Oregon Journalism Project report labeling footage of Washington County commissioner Nafisa Fai a deepfake, condemning it as racist and calling for AI regulation. But testing of the competing images found an OpenAI watermark in the upright-flag image presented to rebut the circulating video.

The findings challenge the factual premise behind those statements. The available evidence does not establish that someone digitally inverted the American flag in the viral footage. Instead, a direct watermark test and visible inconsistencies support the conclusion that the upright image published by OJP contains AI-generated or AI-modified material.

The distinction matters because Democratic officials presented fabrication as established fact while attaching forceful accusations of racism and deliberate harm to the episode.

In an October 19 post, Oregon Democrats wrote:

“The racist deepfake video of Washington County Commissioner Nafisa Fai circulating on social media is abhorrent.”

The party linked OJP’s article and argued that the incident demonstrated a need to regulate AI to prevent harmful misinformation.

U.S. Rep. Maxine Dexter escalated that criticism in an October 20 post:

“This is a racist, disgusting attack and yet another example of how misleading and malignant AI deepfakes can be.”

Dexter praised Fai’s public service, expressed support for her and called for AI regulation. Her post also linked the OJP report.

State Rep. Dacia Grayber similarly described the footage as a deepfake in a Facebook post. The publicly accessible preview called it a “horrific byproduct of unchecked AI and social media algorithms weaponized to foment fear and hate.”

Those statements went beyond defending Fai against hostile commentary. They asserted that the video itself was an AI fabrication, a conclusion the evidence reviewed for this analysis does not substantiate.

Become a Paid Subscriber for Exclusive Content Access and Support Independent News

OJP’s report supplied the basis for the claims

The dispute concerns footage of Fai speaking at a Minnesota community event. Fai is running against fellow commissioner Pam Treece for Washington County board chair.

OJP’s October 19 article categorically described the circulating video as a deepfake. It published an upright-flag screenshot credited to Fai and identified the competing upside-down version as altered.

The report cited attendee Hoodo Mohamed’s recollection that the flag was properly displayed and said OJP received a clip showing it upright. Fai told the publication she did not specifically remember checking the flag’s orientation.

The article does not describe independent forensic testing of either version or explain how it established which version was altered. The version retrieved October 20 continued to carry the deepfake characterization.

Willamette Week published OJP’s reporting, and Axios subsequently repeated its conclusion. That spread the claim but did not provide an additional, independent authentication of the imagery.

Testing found an AI watermark in the rebuttal image

For this analysis, the upright image was downloaded from OJP’s image host and submitted without changing its image content to OpenAI’s verification service.

The result was “Generated with OpenAI tools.” The detailed finding stated: “We found a SynthID watermark that originated from OpenAI.”

Two comparison files were tested separately: a published screenshot of the upside-down footage and a frame extracted approximately 10 seconds into the circulating video. Neither returned an OpenAI provenance signal.

Material tested	Observed result
Upright-flag image published by OJP	OpenAI SynthID watermark detected
Published screenshot of the upside-down footage	No OpenAI provenance signals detected
Separately extracted video frame	No OpenAI provenance signals detected

The positive result is evidence of AI involvement in the published upright image. It does not identify who used the tool, what instructions they supplied or precisely which part of the image was generated or modified.

The negative results also have limits. They do not rule out conventional editing, another provider’s AI tools or processing that degraded a watermark. The test therefore does not certify the entire circulating video as authentic.

The visual comparison points in the same direction

The strongest visible concerns in the upright image occur around the broadcast graphic at the bottom of the frame.

Immediately left of the circular portrait, a red-and-white fragment appears below a straight horizontal boundary. Its position and direction are more consistent with the inverted flag arrangement than with the upright flag above it. Portions visible below the banner also appear to retain the earlier arrangement.

The gold fringe presents a similar discontinuity: its position changes above the graphic while a separate gold strip remains visible beneath the banner farther right.

These features are consistent with replacing the exposed flag while leaving parts around the broadcast graphics unchanged. They are visual observations, separate from the watermark finding. Readers can compare the upright image and the upside-down screenshot.

Multiple frames sampled across the circulating video show a consistent inverted flag arrangement as Fai moves. That supports its plausibility, although it does not substitute for examining the camera-original recording.

The racism accusation does not resolve authenticity

Oregon Democrats and Dexter combined two different judgments: that the episode was racist and that the footage was fabricated.

The original End Wokeness post included the statement, “These are not Americans.” OJP also reported comments calling for Fai’s deportation. That context is relevant to the officials’ denunciations.

But commentary the DPO and these politicians view as offensive does not establish that the underlying footage was digitally manipulated.

The evidentiary problem is the certainty with which the party and elected officials called the video an AI deepfake and used that characterization to support demands for regulation. Their linked source offered a competing image as authentic; that image subsequently tested positive for an OpenAI watermark.

What the evidence supports

These qualitative assessments reflect the combined testing and visual review, not numerical probabilities supplied by a detector:

• Very likely: The published upright image contains material generated or modified with OpenAI tools, supported by the positive SynthID result.
• Likely, with strong supporting evidence: The flag in that image was digitally changed, based on the watermark finding and apparent discontinuities around the broadcast graphic.
• Likely: The upside-down version more accurately depicts the event’s flag, although the original recording is still needed for definitive authentication.
• Not established: The circulating video was digitally altered to invert the flag.
• Undetermined: Who altered the upright image, whether Fai knew about an alteration, and whether the flag’s placement at the event was intentional.

The findings place OJP’s verification process and Democrats’ amplification of its conclusion at the center of the controversy. The strongest technical evidence obtained points to AI involvement in the image used to rebut the video, not affirmative proof that the circulating flag footage was fabricated.

The full recording supplied to OJP, the camera-original footage and an explanation of the upright image’s production would help resolve the remaining questions