What happened

  • An arbitrator found that HHS implemented the January 2025 "Return to In-Person Work" presidential memorandum in a way that violated the parties' collective-bargaining agreement and was an unfair labor practice. HHS filed exceptions on nonfact, essence and exceeded-authority grounds.
  • The Authority denied all three grounds on the merits. The AI discussion sits inside its nonfact analysis and in the footnotes that follow it.
  • One of HHS's authorities for its nonfact argument had a citation the Authority said does not exist ("U.S. Dep't of the Air Force, 913th Air Wing, 57 FLRA 469, 472 (2001)"). The other three authorities, the Authority said, do not concern the legal effect of failing to request bargaining after notice of a change.
  • The Authority then itemized the rest of the brief. Nine more citations were "entirely fictitious," eleven did not support the propositions they were cited for, and ten had partially incorrect citations and did not support their propositions.
  • Its totals: twenty-two of the brief's twenty-nine footnotes, and thirty-five of its forty-two legal or record citations, were fictitious, erroneous or irrelevant.
  • The decision lists the agency's representative as someone in HHS's Human Resources Division. It does not say whether that person is a lawyer, and it does not report an admission that AI was used.

What the court said

The Authority treated the pattern, not any one citation, as the signal: "Based on the nature and scope of these errors, it appears that the Agency may have used a generative artificial-intelligence (AI) tool in drafting its brief" (pp. 7-8).

It then set a rule of thumb for parties in its cases: "nothing prohibits parties from using AI tools when preparing filings for submission to the Authority," but those who do should "be aware that AI continues to present significant risks of error," and should "take steps to verify the accuracy of any AI-generated material" (p. 8).

The consequence was on the arguments, not on the filer. Under 5 C.F.R. § 2425.6, a party has a duty to explain how an award is deficient under the decisional law. The Authority denied one nonfact argument as unsupported because the agency gave no relevant factual or legal support for it. It cited similar warnings from the NLRB (Am. Backflow & Fire Prevention, 375 NLRB No. 31) and GAO (Raven Investigations, B-423447) in its footnotes. It imposed no sanction.

What it means for your firm

  • Agencies and administrative bodies now say this in their own decisions. A federal labor tribunal put the 35-of-42 count in writing and treated bad citations as a reason an argument failed.
  • A non-lawyer party representative was not treated differently from a lawyer on the duty to verify. The Authority, quoting an NLRB decision, says the duty to avoid fabricated citations applies to attorneys and non-attorney representatives alike (p. 8, n.55).
  • Our reading: the most expensive result here was not a fine. It was that a nonfact argument failed as unsupported, because its legal authorities could not carry it.
  • Our reading: if your firm or your client's HR or labor team files exceptions, grievances or briefs before agencies, the verification step has to cover people outside the legal department too.
  • Our reading: opposing counsel who reads the other side's authorities page by page is the cheapest detection layer. The Authority did that work itself here, in footnotes 43 to 52.

What to do Monday

  1. Before any filing goes to an administrative tribunal, have someone other than the drafter pull each cited decision and confirm the party names, reporter page and pincite match.
  2. Add a column to your citation check for "does the decision say what we cite it for," since eleven of the flagged citations here were real decisions cited for propositions they do not support.
  3. Tell non-lawyer representatives who file on your client's behalf, in HR, labor relations or compliance, that the same verification duty applies and who checks their work.
  4. Check the rules of each agency you appear before for an AI statement, and note the date you checked in your filing checklist.

Sources

Questions and answers

Did the Authority find that the agency used AI?

It said the agency "may have used" a generative AI tool, based on the nature and scope of the errors. The decision does not report an admission and does not name a tool.

Was anyone sanctioned?

No. The Authority denied the exceptions on the merits and warned parties about AI risk. It did not impose a fine, strike the brief or refer anyone to a disciplinary body.

How many citations were wrong?

The Authority counted twenty-two of the brief's twenty-nine footnotes and thirty-five of its forty-two legal or record citations as fictitious, erroneous or irrelevant.

Operational information, not legal advice. Facts come from the court record linked under Sources, read on Oct 27, 2026. The reading of what it means for your firm is ours. A source link does not replace the full docket or later history.