The AI Abuse Registry

An AI-upscaled bystander video was offered as evidence in a murder trial and excluded; reported as a possible first-of-its-kind ruling on AI-enhanced video.

In State of Washington v. Puloka (King County, 2024) the defense in a triple-murder case sought to admit an AI-upscaled version of a 10-second bystander smartphone video, processed with Topaz Labs AI and Adobe software. The court excluded it under the Frye general-acceptance standard / ER 403, holding the technique was not generally accepted, used opaque methods to depict what the model 'thought should be shown,' and risked a 'trial within a trial.'

Arrested while eight months pregnant on a facial-recognition match; reported as the first such wrongful arrest involving a pregnant woman.

Porcha Woodruff was arrested by Detroit Police while eight months pregnant on a facial-recognition match. Woodruff v. Oliver (E.D. Mich.) challenged the reliance on the match as failing to establish probable cause — reported as the sixth known wrongful-arrest case tied to the technology and the first involving a pregnant woman.

Tesla's lawyers argued Elon Musk's real recorded Autopilot-safety statements might be AI deepfakes to avoid attribution; the judge rejected it as 'deeply troubling.'

In the wrongful-death suit over Apple engineer Walter Huang (killed in a 2018 Model X Autopilot crash), Tesla's lawyers argued Musk's recorded public statements promoting Autopilot might be AI-generated deepfakes and so unattributable. Judge Evette D. Pennypacker rejected the argument as 'deeply troubling,' warning public figures could otherwise 'hide behind the potential for their recorded statements being a deep fake.' The canonical example of the 'liar's dividend.'

Jailed for nearly a week on a facial-recognition match for thefts in Louisiana, a state he had never set foot in; the warrant rested on the AI match alone.

Randal Reid was jailed for nearly a week in late 2022 after the Jefferson Parish Sheriff's Office identified him via facial recognition for thefts; he had never been to Louisiana, differed from the actual suspect by roughly 40 pounds and a facial mole, and the detective sought the warrant on the AI match alone. He filed suit in 2023.

Wrongfully arrested and detained on a false facial-recognition match; widely reported as the first publicly known US wrongful arrest from the technology.

Robert Williams was arrested by Detroit Police in 2020 on a faulty facial-recognition match — the first publicly known US wrongful arrest from the technology. The ACLU sued on his behalf; the matter was later settled with policy reforms.

Roughly 26,000 families were wrongly accused of childcare-benefit fraud and ordered to repay tens of thousands of euros, with a risk model that used dual nationality as a flag; ruin, bankruptcies, and child removals followed, and the Rutte government resigned.

The toeslagenaffaire saw the Dutch tax authority wrongly brand about 26,000 families as childcare-benefit fraudsters between 2013 and 2019, demanding repayment of tens of thousands of euros; a self-learning risk model used markers including dual nationality. In February 2020 a Dutch court ordered the related SyRI risk-scoring system halted as a human-rights violation. PM Mark Rutte's cabinet resigned in January 2021 over the affair, which a parliamentary inquiry found violated fundamental rule-of-law principles.

Alleged use of AI to create evidential material across a number of cases; reported as the first case of its kind in the UK criminal-justice system.

In June 2026 Derbyshire Constabulary confirmed a criminal investigation into an allegation of perverting the course of justice after the alleged use of AI by a single, unnamed officer to create evidential material in a number of cases. The officer has been removed from frontline duties; no arrests have been made; the officer has not been named. The CPS is working with the force to identify affected cases. This is an ALLEGATION under active investigation — not a proven fabrication, not a force-wide program, not a conviction.

Per EFF, the tool erases the AI draft on export, destroying any record of which report language was AI-written vs officer-written — defeating audit and accountability for reports used in prosecutions.

Axon, the largest US supplier of police body cameras, sells Draft One, which uses a ChatGPT variant to draft police-report narratives from body-camera audio. An EFF investigation found the product 'seems designed to stymie any attempts at auditing, transparency, and accountability': on export it erases the initial draft, so when a report contains 'biased language, inaccuracies, misinterpretations, or lies' there is no way to tell whether the officer or the AI produced it. Audit/transparency criticisms are EFF characterizations.

A Hong Kong finance employee was tricked into 15 wire transfers totaling ~$25.6M after a video call in which every participant except the victim was an AI-generated deepfake of company executives.

In January 2024 a finance employee at the engineering firm Arup in Hong Kong made 15 transfers totaling about HK$200M (~$25.6M) to five accounts after a video conference in which the 'CFO' and other colleagues were all AI-generated deepfakes. Arup confirmed itself as the victim in May 2024; one of the largest documented AI-enabled financial frauds. The Hong Kong police investigation remains open; no arrests announced, funds unrecovered.

Class action alleges an AI tool (nH Predict) with a claimed ~90% reversal-on-appeal error rate was used to override physicians and cut off post-acute care for elderly Medicare Advantage patients; named plaintiff Gene Lokken, 91, was cut off after 19 days and his family paid ~$150,000 before his death.

In November 2023 the Estate of Gene Lokken and others filed a class action (D. Minn.) alleging UnitedHealth and subsidiary NaviHealth used the nH Predict algorithm to deny extended post-acute care to elderly Medicare Advantage members, despite an alleged ~90% error rate (measured by reversals on appeal) and knowledge that only ~0.2% of patients appeal. Lokken, 91, was cut off after 19 of his approved therapy days; his family paid ~$150,000 out of pocket until his death in July 2023. In 2025 a federal magistrate ordered broad discovery into UHC's AI claims process. (Allegations unproven; litigation active.)

More than 20 girls aged 11-17 had AI-generated nude images made from their social-media photos and circulated; a youth court convicted the minors involved.

In September 2023, in Almendralejo, Spain, more than 20 girls aged 11 to 17 had AI-generated nude images created from their Instagram photos using the 'Clothoff' undressing app and shared in WhatsApp groups. A youth court in Badajoz later convicted the minors involved on counts of producing child-abuse imagery and offenses against the victims' moral integrity. An early, widely cited non-consensual AI-imagery harm involving children.

Two days before Slovakia's 2023 election, a deepfake audio of Progressive Slovakia leader Michal Simecka and journalist Monika Todova 'discussing' rigging the vote spread on social media during the pre-election media blackout.

Two days before Slovakia's 30 Sep 2023 parliamentary election, an AI-generated audio clip purporting to capture Progressive Slovakia leader Michal Simecka and Denník N journalist Monika Todova planning to rig the vote (buying Romani votes, manipulating the count) spread on social media during the pre-election quiet period. It was a deepfake; the clip was hard to debunk in time. One of the first EU examples of AI used to manipulate a journalist's and candidate's likeness in an election.

Six wholly fabricated judicial decisions generated by ChatGPT were filed in federal court; the attorneys and firm were fined $5,000 under Rule 11.

In Mata v. Avianca (S.D.N.Y. 1:22-cv-01461), attorneys filed an opposition brief citing six nonexistent judicial decisions generated by ChatGPT, with invented quotes and holdings attributed to real judges. On 22 June 2023 Judge P. Kevin Castel imposed $5,000 in Rule 11 sanctions and ordered the lawyers to notify each falsely identified judge. The canonical AI-hallucinated-citation case; legal researcher Damien Charlotin's database now tracks 1,200-1,600+ such filings worldwide (a continuously growing tally).

ChatGPT fabricated a fake lawsuit stating radio host Mark Walters had embezzled from the Second Amendment Foundation, an organization he never worked for; Walters brought the first defamation suit against a generative-AI product.

In 2023 a journalist, Fred Riehl, asked ChatGPT to summarize a real lawsuit; the model invented a different one, reporting that 'Armed America Radio' host Mark Walters had embezzled from the Second Amendment Foundation. None of it occurred — Walters never worked for the SAF and was never accused of any such thing. Walters sued OpenAI, the first defamation suit against a generative-AI product. On 19 May 2025 Judge Tracie Carson (Gwinnett County Superior Court) granted OpenAI summary judgment, holding the output was not defamatory as a matter of law because no reasonable reader would have taken it as a statement of fact — the journalist knew ChatGPT fabricates, saw the warnings, and could be expected to verify. The fabrication is proven; the legal outcome was a defense win. Walters was the subject of the fabrication, not a wrongdoer; the embezzlement never happened.

An Arizona mother was nearly extorted for ransom after scammers played what sounded exactly like her daughter sobbing and a kidnapper demanding money; she testified to the US Senate.

In April 2023 Jennifer DeStefano received a call in which she heard what sounded exactly like her 15-year-old daughter Brianna sobbing, followed by a man demanding ransom (initially $1M, then $50,000). The voice was an AI clone; her daughter was safe at home. DeStefano testified about the scam before the Senate Judiciary Committee on 13 June 2023. A widely documented example of voice-clone extortion.