Justice Department finds misconduct by two lawyers who used generative AI


The US Department of Justice's Office of Professional Responsibility has concluded that professional conduct rules were violated in two separate incidents involving government lawyers using generative AI technologies. The incidents included submitting a court brief containing fabricated citations while making misleading statements to the judge, and uploading judicially restricted work documents to a public AI platform.
Documented internal oversight findings at the US Department of Justice have revealed two separate incidents in which government lawyers violated professional conduct rules while using generative AI tools, placing standards for legal verification and the protection of restricted document confidentiality at the forefront of judicial governance.
The US Department of Justice’s Office of Professional Responsibility described two separate incidents in which government lawyers breached professional conduct requirements when using generative AI technologies. Reuters was the first to report details of these two incidents on November 18, 2026, at 17:00 UTC (21:00 Gulf Standard Time), following the office’s release of official summary findings the previous day, November 17.
These reports, also published via the Westlaw Today platform, are particularly significant because they involve documented internal investigation findings proving actual professional misconduct, rather than being limited to fleeting allegations that an AI model produced an incorrect answer.
In the first incident, an assistant US attorney submitted a court document containing fabricated citations and a distorted presentation of legal rulings and principles. When the judge questioned the attorney about these issues, the attorney initially stated that a preliminary draft had been filed by mistake instead of openly and transparently disclosing the use of AI in preparing the document.
The Office of Professional Responsibility concluded that the attorney made false or misleading statements regarding how those errors originated. Under established legal practice, no citation generated by a language model with a confident tone holds any evidentiary value unless it literally and substantively matches the original legal authority, as lawyers remain fully responsible for verifying all court briefs regardless of the software used to assist in their preparation.
In the second incident, another lawyer uploaded work-related documents to a public AI platform despite a court-ordered restriction prohibiting the sharing of that information, and in violation of approved internal Department of Justice policies. The office described this action as reckless conduct, while the name of the public platform used in the incident was not disclosed.
Legal and technical analysis shows that uploading any official document to a public external service can create severe confidentiality and access-control issues, even in cases where the generated answer is entirely accurate. The severity of these risks depends on the nature of the uploaded information, existing court orders regarding it, and the technical safeguards applied by the external service.
Both findings from the Office of Professional Responsibility were referred to the department’s Professional Misconduct Review Unit, the body authorized to determine appropriate consequences and disciplinary penalties. The US Department of Justice did not disclose the names of the lawyers involved, their geographic work locations, or their current employment status within the department.
It also remains unclear whether the department reported these two incidents to legal disciplinary bodies and state bar associations. Those details remain undisclosed.
Generative AI can be a useful tool for organizing documents and identifying research questions; however, it can produce invented legal citations, rewritten phrasing presented as direct quotes, or inaccurate descriptions of court rulings, making direct verification of original sources by lawyers mandatory. Government agencies also require strict rules for handling protected or restricted information, along with training users on the fundamental difference between approved internal systems and open public services.
The significance of these two cases extends beyond individual incidents, as professional governance must address both accuracy and confidentiality as twin pillars. Future developments will involve any disciplinary decisions issued, updates to department guidance, and the publication of approved verification standards, noting that the prior case involving a prosecutor in North Carolina referenced by Reuters is part of the historical background and does not represent a third incident on November 18.
Question: What is the nature of the two violations identified by the US Department of Justice’s Office of Professional Responsibility?
Answer: The first incident involved submitting a court brief containing fabricated citations and misleading statements to the judge regarding the cause of the error, while the second involved uploading documents restricted by a court order to a public AI platform.
Question: To which entity were the internal investigation findings referred to take necessary action?
Answer: The findings in both cases were referred to the Department of Justice’s Professional Misconduct Review Unit to determine the resulting disciplinary consequences.
Question: Did the Department of Justice disclose the names of the lawyers or the name of the AI platform used?
Answer: No, the department did not disclose the names of the lawyers, their locations, or their current employment status, nor did it specify the name of the public AI platform in the published summaries.
Question: Is the case involving a prosecutor in North Carolina part of the new November 18 incidents?
Answer: No, the prior North Carolina case was cited by Reuters for contextual background only and is not a third incident among the current November summaries.