How employers use Artificial Intelligence (AI), and the documentation of that use, is impacting Employment Practices Liability (EPL). Prosecutors now ask whether companies risk-assessed AI and built controls; judges are requiring disclosure of AI use; and early cases show courts will test whether “human in the loop” is real or symbolic.
Galloway employment lawyers offer their insights to the growing trend that is shaping EPL practice and litigation.
Use AI for speed, not judgment—keep humans accountable, document rigorously, validate for bias, and align with fast-moving enforcement across the U.S. and EU.
What changed, and why does it matter now?
Wendell Hall: “As the Louisiana 5th Circuit recently reminded us in In re Kenney, the court won’t sanction the software—it will sanction the lawyer. Accuracy is a human responsibility, not a software feature.”
The point is simple: you must verify AI outputs; courts view unverified reliance as “objectively unreasonable.”
The Department of Justice’s September 2024 update to the Evaluation of Corporate Compliance Programs (ECCP) added AI risk management. With prosecutors now asking whether companies risk‑assessed AI and built controls, federal judges have issued standing orders requiring disclosure of AI use. Ongoing cases (Mobley v. Workday; Kistler v. Eightfold) test ADEA and FCRA theories against AI screening.
What is in scope?
Autumn George: “Carriers are not just asking whether AI was used in hiring. They are increasingly focused on whether AI touched any part of the employment lifecycle: recruiting, screening, performance management, productivity monitoring, investigations, discipline, terminations, reductions in force, leave management, accommodation decisions, and workplace surveillance.”
“From a claims perspective, the key question is not simply, ‘Was AI used?’ It is, ‘Did AI influence the challenged decision?’”
U.S. federal laws (EEOC/Title VII, ADEA, ADA, FCRA) remain in play, in addition to state biometric regimes. In particular, industries such as healthcare, retail, technology, and gig platforms face heightened risk where AI touches hiring, monitoring, and termination decisions.
How should employers use AI in investigations without increasing liability?
George: “If AI was used to summarize evidence, flag communications, or assist with analysis, the file should reflect what tool was used, what information was reviewed, whether the output was verified, and who made the final decision.”
“Employers create unnecessary risk when the file suggests reliance on unexplained AI conclusions or does not show independent human review of the underlying facts.”
- Require meaningful human-in-the-loop for interviews, findings, and credibility determinations; treat AI outputs as drafts and verify before they enter the evidentiary record.
- Preserve prompts, inputs/outputs, model versions, retrieval sources, and reviewer sign‑offs; align with judge-specific AI disclosure rules.
- Bar public LLMs for privileged/PII; enforce vendor transparency, audit rights, and bias/accuracy testing pre-deployment.
With a litigation lens, is AI a helpful tool or a defensibility challenge?
Doris Bobadilla: “Both—AI can help your defense team and can hurt the client’s defensibility. To prepare the defense, AI assists in the organization, which should save significant time. As the saying goes: junk in, junk out—if outputs are accurate and well-documented, AI may strengthen the process. If not, it creates credibility gaps that opposing counsel will exploit. In the end, it is your client’s process that will be judged.”
“AI can make preliminary, routine steps much easier to do, which is no doubt alluring as it may save tons of labor hours, producing excellent tools that lawyers can use to better the client’s interests and sharpen their defense. Yet it often feels like ‘Cliff Notes,’ where relying solely on these summaries can leave you missing the nuances of the story. What keeps me up at night is that clients might overlook the story’s essence when using these tempting tools.”
For employers, what are your thoughts about the importance of training?
Hall: “Training isn’t just about how to use the tool; it’s about how to audit it. Proactive training should include ‘Red Teaming,’ that is, having humans intentionally challenge the AI’s conclusions to ensure the ‘logic’ holds up under cross-examination.”
Francis Waguespack: “User knowledge of the program, how to search and utilize prompts, and privacy concerns and security features” are essential elements of training. Plus, enforce policy acknowledgments and progressive discipline: “If the policy is violated… suspension of AI access… additional security training… [and] termination may be appropriate” for repeat violations.
Bobadilla: “Like many advances, employers should build clear guardrails and train their teams on the appropriate use of AI. What you often see—understanding that employers are always busy with their business mission—is that they don’t train their employees proactively. Instead of reacting to a mishap or worse, proactive training should be a constant focus and can also later be used if needed as a defense mechanism.”
Where are the legal exposures and how are enforcers approaching AI?
Hall: “The EEOC’s iTutorGroup settlement shows age‑based auto‑rejection by AI triggers liability under the ADEA; and In re Kenney confirms courts will sanction humans for unverified AI work product.”
“Treat AI as your most industrious junior associate… You can outsource the labor to an algorithm, but you can never outsource the liability.”
How does AI affect settlement strategy?
George: “In reputation-sensitive cases, the mere allegation that an employer used a ‘black box’ AI tool can create business risk beyond the legal merits… If the employer cannot explain how the tool worked… whether the output was validated, or who made the final decision, the AI issue can become a settlement driver. Defensible use must be built before the claim.”
What should EPL leaders implement now?
Waguespack: “Short answer—an active role for legal in approving/auditing HR-facing AI, plus signed policy acknowledgments, training, and graduated consequences.”
- Map AI across HR decisions; ban AI from making credibility or disciplinary determinations; require human sign‑off and rationale independent of the tool.
- Update SOPs: capture tool used, preserve inputs/outputs, validation steps, and decision-makers; align with ECCP documentation expectations.
- Amend vendor contracts/Data Processing Agreements: transparency, audit rights, bias/accuracy testing; restrict public LLMs for PII/privileged data; preserve audit trails.
- Train managers on “red teaming,” prompt hygiene, and documentation discipline; require acknowledgments; enforce consequences for misuse.
What are insurers asking and how should underwriters respond?
George: “Underwriters should move beyond the threshold question of whether the insured ‘uses AI’… The better questions focus on where AI is being used, who approved it… whether the employer has policies requiring human review, vendor oversight, bias testing… data retention… and legal escalation for high-risk uses.”
“The greatest risk is not necessarily the company using AI. It is the company using AI without knowing where, how, why, or by whom it is being used.”
Key Takeaways
AI doesn’t create the liability—the failure to review it does. Consistent, documented human verification and explainability across cases—paired with bias testing and audit trails—are what keep investigations defensible and underwriting realistic.
- “Meaningful human involvement” must be real; symbolic review collapses in court.
- Document the role AI played, what was verified, and who decided—before the claim arrives.
- Prosecutors and judges expect AI risk assessment, disclosure, and controls; build ECCP-aligned governance now.
- Treat AI outputs as drafts; preserve prompts, inputs/outputs, and sign‑offs; bar public LLMs for privileged/PII.
- Train to audit: “Red Team” AI conclusions and enforce policy compliance with graduated discipline.
Disclaimer: This material is provided for informational purposes only. It is not intended to constitute legal advice, nor does it create a client-lawyer relationship between Galloway and any recipient. Recipients should consult with counsel before taking any action based on the information contained within this material. This material may be considered attorney advertising in some jurisdictions.
