90 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 13 July 2026
← The incident library
HD-INC-036
Legal services · United States · 2024 · Hallucination & fabrication

A misinformation expert filed court testimony defending a deepfake law, and his AI-generated citations pointed to studies that did not exist

By Ellie Harris · Filed 1 November 2024 (declaration filed)

Alleged: Stanford University (expert witness); Office of the Minnesota Attorney General (retaining party) developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A misinformation expert filed court testimony defending a deepfake law, and his AI-generated citations pointed to studies that did not exist

What happened

The case, Kohls v. Ellison, was a challenge to Minnesota’s 2023 law restricting the use of deepfakes to influence an election, brought by a conservative satirist and a state representative. The Minnesota Attorney General was defending the law, and engaged Jeff Hancock, a Stanford communication professor and a well-known scholar of misinformation, as an expert. On 1 November 2024 Hancock filed a twelve-page declaration arguing that deepfakes amplify misinformation and erode trust in democratic institutions.

The declaration cited a number of sources. At least two of them could not be found, because they did not exist, and a third misattributed its authors. Hancock had used GPT-4o, alongside Google Scholar, to help assemble the citation list, and had not verified the entries before signing the declaration under oath. The opposing lawyers could not locate the studies and brought it to the court’s attention. Hancock acknowledged that the citations were AI-generated and unchecked, and wrote that he had not intended to mislead the court or counsel. In an order dated 10 January 2025, Judge Laura Provinzino excluded the declaration, writing that his citation to fake, AI-generated sources “shatters his credibility with this Court.” She also denied the Attorney General leave to file an amended version, so the exclusion was final rather than a setback the state could repair.

The irony was noted widely, and it is the point. A recognised authority on AI-driven misinformation had filed, in defence of a law against AI-driven deception, citations that an AI had fabricated. The judge made the point herself, writing that Hancock had “fallen victim to the siren call of relying too heavily on AI” in a case about the dangers of AI. Expertise in the subject did not protect him, because the missing step was not knowledge. It was verification.

What an auditable version would have shown

The professional duty in a sworn filing is old and unambiguous: the authorities you cite must be real and must say what you claim. What was missing was any record connecting each citation to a confirmed source. An auditable version logs every citation as it is assembled, with a status of verified, partially verified, or unverified and the source consulted, and checks the final document against that log before it is filed. A citation still marked unverified blocks the filing. With that record, the two fabricated entries would have surfaced at the desk, before the declaration was signed, rather than in the opposing brief.

Where the gap was

The gap is the same one running through the legal cases in this library: a generative tool returns plausible, correctly formatted citations whether or not the underlying work exists, and the lookup that used to guarantee a real source has been removed without anyone noticing. The control is a CitationVerifier: every citation produced with AI assistance is checked against a primary source before the document leaves the desk, and an unverifiable citation stops the document. A ConductRecord preserves the prompts, the model version, the citations, and the verification result, so a practitioner can show how each authority was confirmed.

What governance should have looked like

The defence is not that experts should avoid AI. It is a verification gate built into the work, that no filing passes without, plus the duty to disclose AI use and to be able to produce the record of how citations were checked. That record is what restores the assumption a court has always relied on, that a cited authority has been confirmed against a real source by someone. The signed record is what makes the disclosure verifiable rather than a promise.

The reference implementation of CitationVerifier and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Stanford University (expert witness); Office of the Minnesota Attorney General (retaining party) could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →