In the fast-evolving world of public sector labor law, tools like generative AI promise efficiency and innovation. But as a recent decision from the Public Employment Relations Board (“PERB”) reminds us, they can also spell disaster if not handled with the utmost care. In California State University Employees Union v. Trustees of the California State University (San Diego) (PERB Case No. LA-CE-1433-H), an Administrative Law Judge (“ALJ”) took the extraordinary step of striking the employer’s pre-hearing brief from the record. The reason? Fabricated citations and quotations from a federal appellate decision that simply didn’t exist as presented—hallmarks of unchecked AI output.
Let’s break this down step by step, because this isn’t just a procedural hiccup; it’s a wake-up call for unions, employers, and practitioners alike in California’s public safety sector.
The Case Background
The underlying dispute centers on the employee status of residential assistants (“RAs”) under the Higher Education Employer-Employee Relations Act (“HEERA”), specifically Government Code section 3562(e). The union argued that student RAs qualify as employees entitled to bargaining rights, while the Trustees of the California State University (“CSU”) contended otherwise. In preparation for a hearing, the ALJ directed both parties to submit pre-hearing briefs addressing the legal test for employee status and the relevance of federal Fair Labor Standards Act (“FLSA”) precedents.
CSU filed its brief on November 3, 2025, relying heavily on Marshall v. Regis Educational Corp. (10th Cir. 1981) 666 F.2d 1324 (“Marshall”)—a real case, but one that CSU misrepresented through inaccurate page citations and invented quotations. For instance, the brief claimed the Tenth Circuit held that RAs’ duties were “primarily educational rather than economic in nature” and that they “receive the greater benefit from the program.” In reality, Marshall ends at page 1328 and contains none of these phrases or conclusions. The ALJ spotted the discrepancies, issued an Order to Show Cause, and ultimately struck the entire brief as a sanction after CSU’s response failed to adequately explain the errors.
CSU admitted to “misnumbering of pages” and “erroneously included quotation marks around paraphrasing statements,” attributing it to a failure to double-check. But the ALJ wasn’t buying it, noting that the misrepresentations went beyond mere typos—they distorted the case’s holdings in a way that aligned suspiciously with CSU’s position. Drawing parallels to Noland v. Land of the Free, L.P. (2025) 336 Cal.Rptr.3d 897, where a California appellate court sanctioned counsel for AI-generated fabrications, the ALJ emphasized that such “hallucinations” undermine the integrity of legal proceedings.
Why This Matters for Public Safety Unions
For unions representing California’s firefighters, police officers, corrections staff, and other public safety workers, this ruling underscores a critical lesson: diligence in legal advocacy isn’t optional. PERB proceedings, like those under the Meyers-Milias-Brown Act (“MMBA”) or the Dills Act, demand precision because the stakes—bargaining rights, working conditions, and member protections—are high. Imagine a grievance over shift differentials or safety equipment where a union’s brief gets tossed due to sloppy AI use. Not only does it weaken your position, but it could invite scrutiny or countersanctions that distract from the merits.
More broadly, this decision signals PERB’s intolerance for shortcuts in an era where AI tools like ChatGPT are tempting for drafting research summaries or arguments. As the ALJ pointed out, citing CSU’s own AI guidelines, it’s the attorney’s responsibility to verify content. In public sector labor, where decisions often set precedents affecting thousands of members, relying on unvetted AI could erode trust with arbitrators, boards, or courts. We’ve seen similar pitfalls in federal cases, but this is one of the first in California’s public employment arena—and it happened to a major employer like CSU, which should know better.
The potential ripple effects? Expect heightened scrutiny of briefs in PERB and related forums. Unions might see employers trying to exploit AI for aggressive positions, only to backfire as in this case. On the flip side, it empowers unions to challenge dubious citations, turning the tables in discovery or hearings. And for ongoing debates like student employee status—relevant if your union deals with campus safety personnel—this ruling keeps the focus on statutory language over manufactured precedents.
Lessons Learned and Best Practices
CSU’s misstep highlights how overreliance on AI technology
can backfire. AI is a tool that must be used carefully. It’s a substitute
for human research and must be verified. Always cross-check citations, quotes,
and summaries against primary sources. Tools like Westlaw or Lexis are
irreplaceable for this. Follow the State Bar’s guidance on AI, which stresses
competence and candor. Don’t let “enhancements” from AI platforms slip through
without review. Labor organizations, and lawyers alike, should establish
protocols for AI use, including safeguards for confidential information and
verification policies. If you spot hallucinations in an opponent’s filing,
don’t hesitate to call it as doing so will strengthen your case.
