Matt Gaetz’s antifa-detecting facial recognition story is complete nonsense

Yesterday, a mob of Trump supporters overran the US Capitol, causing widespread chaos in an attempt to nullify the results of the 2020 presidential election. In the wake of the attacks, several Republican politicians have claimed the attackers were anti-fascist activists, in spite of the widespread Trump paraphernalia and triumphant social media posts by Trump supporters. But there’s no evidence antifa played a notable role in the riot, and one of the most widely cited examples has already fallen apart.

In a widely heard House speech on Wednesday, Rep. Matt Gaetz (one of 147 Republican Congress members who voted to overturn the 2020 presidential election results) claimed that the mob had been infiltrated by antifa. But Gaetz cited confusing, unverifiable facial recognition evidence from a company that now calls the original article defamatory — and says it identified neo-Nazis, not antifa supporters.

In a speech during the process of certifying President-elect Joe Biden, Gaetz claimed there was “some pretty compelling evidence from a facial recognition company” that some Capitol rioters were actually “members of the violent terrorist group antifa.” (Antifa is not a single defined group, does not have an official membership, and has not been designated a terrorist organization, although President Donald Trump has described it as one.)

Gaetz attributed this claim to a short Washington Times article published yesterday. That article, in turn, cited a “retired military officer.” The officer asserted that a company called XRVision “used its software to do facial recognition of protesters and matched two Philadelphia antifa members to two men inside the Senate.” The Times said it had been given a copy of the photo match, but it didn’t publish the picture.

There is no evidence to support the Times’ article, however. An XRVision spokesperson linked The Verge to a blog post by CTO Yaacov Apelbaum, denying its claims and calling the story “outright false, misleading, and defamatory.” (Speech delivered during congressional debate, such as Gaetz’s, is protected from defamation claims.) The Times article was apparently deleted a few hours after Apelbaum’s post.

Image: XRVision

“XRVision didn’t generate any composites or detections for the Washington Times or for any ‘retired military officer,’ nor did it authorize them to make any such claims or representations,” Apelbaum wrote. According to his post, XRVision did analyze video footage of the riots, and the company identified “several individuals” in a composite it shared with a “handful” of outsiders. However, they were not linked with antifa.

We concluded that two of the individuals (Jason Tankersley and Matthew Heimbach) were affiliated with the Maryland Skinheads and the National Socialist Movements. These two are known Nazi organizations; they are not Antifa. The third individual identified (Jake Angeli) is an actor with some QAnon promotion history. Again, no Antifa identification was made for him either.

Angeli, who frequently appears at protests in a horned helmet and face paint, is known as the “Q Shaman” and is affiliated with the conspiracy movement QAnon. Angeli previously participated in a documentary called “The Patriots,” in which he espoused an extreme pro-Trump ideology rather than anything aligned with antifa.

These names tally with earlier evidence posted by critics of the Times piece. The Twitter account Respectable Lawyer, for instance, posted a long thread debunking the claims of antifa involvement. That account noted that Tankersley and Heimbach’s photos did appear on a Philadelphia antifa site, but only because the site was identifying them as neo-Nazis. However, while that thread identified Tankersley by his tattoos, it did not definitively place Heimbach at the riot.

Even after Apelbaum’s update, it’s not actually clear if XRVision’s technology works or how statistics like the “match rate” on the composite were calculated. XRVision’s website offers little information about its software. In a slideshow from a 2019 Nvidia AI Innovation Day presentation in Singapore, XRVision suggested it could perform advanced facial recognition and complex computer vision analysis on a security camera or smart device footage. As OneZero notes, however, the company has apparently not submitted algorithms for testing by the US National Institute of Standards and Technology. Apelbaum posted his statement after activists and journalists had already identified Angeli and Tankersley.

At best, facial recognition remains a flawed technology that can easily misidentify targets. Amazon’s Rekognition system, which was used by US law enforcement until 2020, erroneously matched 28 members of Congress with criminal mugshots. Last month, a New Jersey man sued after being falsely identified and arrested based on an incorrect facial recognition match.

The Times article was apparently deleted without a correction after Gaetz already widely spread its thinly sourced myth about antifa-identifying facial recognition tech. And it’s part of a much broader false theory that “antifa infiltrators” caused the widespread chaos during yesterday’s riot. Texas Attorney General Ken Paxton quoted a tweet speciously claiming a “bus load” of “antifa thugs” had infiltrated the demonstrations, stating that “these are not Trump supporters.” In reality, “antifa buses” are a well-known hoax that led one group of Washington townspeople to terrorize a family on a camping trip.

As The New York Times notes, there is no evidence that antifa or other left-wing figures had a substantial presence at the riot. Washington Times author Rowan Scarborough did not immediately respond to a request for comment, nor did the office of Rep. Gaetz.