ChatGPT-Using Lawyer Sanctioned for Citing Fabricated Witness Testimony
A lawyer has been sanctioned after submitting court filings that cited testimony from witnesses who do not exist — testimony generated by ChatGPT without any verification before submission. The case follows a pattern that has now become familiar enough to have a defined profile: a legal professional uses a generative AI model to assist with research or drafting, accepts its output without independent verification, and files documents that contain fabricated citations, cases, or — in this instance — witness accounts.
What distinguishes this case from earlier incidents involving hallucinated case law is the nature of the fabrication. Fake legal citations are damaging. Fake witness testimony introduces a different category of problem — one that touches on evidence integrity and the foundational reliability of the factual record presented to a court.
The incident surfaces at a moment when courts across jurisdictions are still developing formal policies on AI-assisted legal work. Some have introduced disclosure requirements. Others have issued standing orders requiring attorneys to certify that AI-generated content has been independently verified. Enforcement, however, remains inconsistent, and cases like this one demonstrate that the consequences of that inconsistency are not theoretical.
At the operational level, what happened here reflects a structural failure in how AI tools are being integrated into professional workflows. Generative models like ChatGPT are designed to produce fluent, contextually plausible text. They are not designed to retrieve or verify facts. When used in a research or drafting context without a verification layer — human or automated — the model's outputs carry the surface appearance of accuracy without the underlying integrity that legal and evidentiary standards require.
This is not a failure mode that only affects low-stakes documents. Hallucinated content can propagate into sworn filings, judicial records, and legal arguments that shape real outcomes. The professional disciplinary system provides one check. Courts imposing sanctions provides another. But neither operates at the speed or scale at which AI-assisted work is now being produced across legal practices of every size.
The implications extend beyond the legal sector. Any professional domain where AI-assisted outputs are submitted to an authoritative body — regulatory agencies, audit committees, institutional review boards — faces a structurally similar risk. The friction that once existed between drafting and filing has been compressed by AI tools. That compression is operationally useful, but it also removes stages of the process where errors would historically have been caught.
From an institutional perspective, this case reinforces a position that is becoming difficult to dispute: the adoption of AI tools in high-stakes professional contexts requires explicit workflow controls, not just individual practitioner judgment. Law firms, corporate legal departments, and solo practitioners alike are operating under professional obligations that AI systems have no capacity to fulfill on their own. The model does not know what a court requires. The model does not know what constitutes evidence. The model generates plausible output and stops there.
The lawyer's sanction in this case is a professional consequence. The longer-term consequence is likely procedural — more courts, more bar associations, and more institutional clients will move toward requiring documented verification protocols for any AI-assisted filing. What has been voluntary guidance in many jurisdictions is becoming, case by case, the expected standard of competent practice.
Sources: — Ars Technica (https://arstechnica.com/tech-policy/2026/09/chatgpt-using-lawyer-punished-for-citing-fake-testimony-from-made-up-witnesses/)