Behind the Privilege Shield: Safeguarding AI Bias-Testing Data in Employment Decisions

In this piece for Norton Rose Fulbright, authors Jesika Silva Blanco, Susana Medeiros & Susan Linda Ross look into a significant California federal court ruling in Mobley v. Workday, which held that certain AI bias-testing materials may be protected by attorney-client privilege. The case centers on allegations that Workday’s AI-powered hiring tools discriminated against job applicants, while also highlighting the growing regulatory landscape governing automated employment decision-making, including compliance requirements under New York City Local Law 144.

The authors explain that although Workday was required to produce the underlying historical data used in its bias audits, it argued that the bias-testing analyses themselves were created under the direction of legal counsel to obtain legal advice on regulatory compliance. The court agreed, finding that the testing had been meaningfully directed by attorneys rather than conducted solely for business purposes, and therefore qualified for attorney-client privilege despite being relevant to the plaintiffs’ discrimination claims.

The article emphasizes that the ruling does not extend privilege to raw factual data or reports submitted to regulators, but instead protects legal analyses where counsel plays a substantive role in designing, curating, and overseeing the testing process. The decision suggests that privilege could be lost if testing is performed primarily for business purposes, delegated to outside consultants, or disclosed to third parties or regulatory agencies.

The authors conclude that the decision offers valuable guidance for employers as AI-related employment regulations continue to expand across the United States. Organizations should involve legal counsel early in AI governance and bias-testing efforts, carefully document that testing is conducted to obtain legal advice, limit unnecessary external disclosure of privileged analyses, and recognize that while underlying data remains discoverable, a well-structured compliance program can help preserve privilege over legal evaluations of AI systems…

CONTINUE READING…