Privilege Wins the Round, But the AI Discovery War Just Got Messier
Key quote:
Here, Workday has represented that its attorneys curated the data it used in the bias testing, the overall purpose of the testing was to provide legal advice and not to be used in a business capacity, and it has not submitted the data to a regulatory body. The plaintiffs’ assertions that Workday’s purpose was business or that the underlying code is technical in nature does not change the result. Workday has shown more than mere direction by its attorneys. Thus, the bias-testing data is privileged.
Why it matters:
Magistrate Judge Laurel Beeler’s May 29, 2026 order (PDF) in Mobley v. Workday (Case No. 23-CV-00770) handed a tentative victory to ADMT (automated decision making technology) vendors, but don’t mistake a discovery win for a case dismissal. The court blocked plaintiffs from seeing Workday’s actual bias-testing code and results, ruling that because lawyers “curated” the data for legal advice, it stays shielded behind attorney-client privilege. I’m not sure that’s what folks have in mind with the so-called “human in the loop”, but if this potential expansion of privilege isn’t overturned or appealed, companies might consider this as a strategy for protecting their AI training data from discovery, even though end user prompts and outputs generally seem to be going the same way as Slack messages.
But there’s a catch. While the court hid the internal tests, it forced Workday to hand over its own EEO-1 and OFCCP filings. Why? Because Workday uses the same algorithms to hire its own staff that it sells to clients. The court reasoned that if your internal hiring data shows demographic skew, you probably knew your product had the same flaw.
The ruling also shut down the request to access customer data. Even though Workday’s contracts allow disclosure “to the extent required by Law,” the judge said this didn’t count as “control” under Rule 34 because customers could sue to stop the leak. This forces plaintiffs to chase down thousands of individual employers instead of the platform, which would be an insurmountable logistical nightmare for people who are alleging algorithmic discrimination.
This case still isn’t over. The agency liability theory survived a motion to dismiss, meaning Workday still faces the possibility of being on the hook for millions of rejected applicants. But for now, the code and test results that might have led to biased outcomes remain shielded. Given what Connecticut’s been up to lately, we may see novel litigants trying similar cases under SB5.
