top of page
Search

Artificial Intelligence in Workplace Investigations: What Employers Should Do Before Software Touches the Evidence

  • Todd Nurick
  • 1 day ago
  • 8 min read

Business attorney and HR professionals reviewing workplace investigation evidence and AI compliance risks
Business attorney and HR professionals reviewing workplace investigation evidence and AI compliance risks

Artificial Intelligence, (AI), in workplace investigations can be useful, but it shouldn’t be treated like a shortcut around judgment.


Internal investigations are already sensitive. A company may be dealing with harassment, discrimination, retaliation, theft, threats, leave abuse, wage issues, policy violations, or a broken relationship between employees. The investigator has to collect facts, assess credibility, protect privacy, preserve evidence, avoid retaliation, and make findings that may later be challenged by a lawyer, agency, court, arbitrator, or unemployment referee.


AI can help with some of that work. It can organize documents, search emails, build timelines, summarize interview notes, flag inconsistencies, identify missing evidence, and help draft a cleaner investigation report. Used carefully, it can save time and help an investigator see patterns that might otherwise be missed.


Used carelessly, it can create a different problem. The company may end up with biased outputs, privacy exposure, privilege questions, data-security concerns, unsupported findings, or a disciplinary decision that looks like it was made by software instead of by a trained human decision-maker.


Todd Nurick of Nurick Law Group, LLC is a Pennsylvania and New York business attorney with approximately 30 years of civilian business law and litigation experience and a former Army officer. As Fractional General Counsel and Outside General Counsel, he helps businesses address contracts, employment issues, internal investigations, compliance, risk management, and the full range of legal concerns that an internal legal department would ordinarily handle.


Artificial Intelligence in Workplace Investigations can help, but only if the company controls the use case

The first question isn’t whether AI can be used in an investigation, because it obviously can. The better question is what the company is asking AI to do. There’s a major difference between using AI to help organize a large file and using AI to decide whether an employee is lying. There’s also a major difference between using AI to summarize interview notes and using AI to recommend termination.


In a workplace investigation, AI may be appropriate for tasks such as:

  • sorting large volumes of emails, messages, or documents;

  • identifying potentially relevant communications;

  • building a first-draft chronology;

  • summarizing witness interviews;

  • comparing statements for inconsistencies;

  • identifying documents that may need human review;

  • helping format a final report after the investigator has made the findings.


Those are support functions but, as you may have surmised, they don’t replace the investigator. The Equal Employment Opportunity Commission, (EEOC), has made clear that federal employment-discrimination laws still apply when employers use automated systems, including AI, to make or inform employment decisions. The Department of Labor, (DOL), has also emphasized meaningful human oversight, transparency, protection of worker data, and respect for labor and employment rights when AI is used in the workplace.


Artificial Intelligence in Workplace Investigations shouldn’t become a credibility machine

One of the most dangerous uses of AI in an investigation is “truth detection.” Some tools claim to assess emotion, stress, deception, tone, eye movement, facial expression, voice patterns, or body language. That may sound attractive when witnesses disagree, but it’s exactly the kind of area where legal risk can outrun the technology.


The Employee Polygraph Protection Act, (EPPA), generally prohibits most private employers from requiring, requesting, suggesting, or causing an employee or job applicant to take a lie detector test. The DOL also says employers generally may not use, accept, refer to, or inquire about lie-detector results, and may not discipline or discriminate against an employee for refusing to take such a test.


That doesn’t mean every AI interview tool is automatically a lie detector. It does mean employers should be very careful before using AI to evaluate truthfulness, deception, emotional state, or credibility during an investigation. A trained investigator can consider demeanor as one piece of a broader record. A company shouldn’t outsource credibility findings to software that may be unreliable, biased, poorly validated, or legally questionable.


The discrimination risk doesn’t disappear because the tool is “neutral”

AI tools often sound neutral because they use data, scores, categories, rankings, or patterns. That doesn’t mean the result is neutral. The EEOC has warned that Title VII of the Civil Rights Act of 1964 can apply when automated systems are used to make or inform selection decisions, and the agency has discussed how employers may need to assess whether those tools create adverse impact based on protected characteristics. The EEOC has also identified disability-related risks when AI or algorithmic tools screen out individuals with disabilities, fail to account for reasonable accommodations, or require disability-related information in a way that creates compliance concerns.  


That concern carries over into investigations. An AI tool used to review complaints, rank witness credibility, detect “aggressive” language, flag policy violations, or recommend discipline could perform differently across groups depending on the data used, the assumptions built into the model, and the way the tool is deployed.


The National Institute of Standards and Technology, (NIST), has described AI bias as a socio-technical problem, not just a coding problem. NIST’s work notes that bias can be introduced or amplified through data, design, development, deployment, and the larger organizational setting in which the system is used.


That’s the point I think employers should focus on. AI risk isn’t limited to the vendor’s model. It also depends on how the employer uses the tool, who reviews the output, what data goes into it, what decision follows, and whether the company can explain its process later.


Vendor responsibility doesn’t eliminate employer responsibility

Employers sometimes assume the vendor owns the risk because the vendor built the tool-that’s not a safe assumption. The EEOC has stated that Title VII applies to the use of automated systems, including AI, when those systems make or inform employment decisions. The agency’s public materials also make clear that employers need to understand how these tools are being used, not merely accept the vendor’s marketing language.


From a general-counsel standpoint, that means the vendor review matters before the investigation tool is used. The contract should address confidentiality, data retention, model training, security, audit rights, subcontractors, indemnification, cooperation with subpoenas or agency requests, and what happens to employee data after the matter ends.


An employer using AI in investigations should also know whether the tool:

  • stores prompts, uploads, transcripts, or outputs;

  • uses employer data to train or improve the model;

  • permits the vendor or subcontractors to access employee information;

  • creates logs that may be discoverable;

  • can produce an audit trail showing how outputs were generated;

  • has been tested for bias, accuracy, and security;

  • can segregate legally privileged or confidential investigation materials.


Privacy, privilege, and discovery problems can start with one upload

Workplace investigations often involve sensitive information: medical details, complaints about protected activity, wage records, disability accommodations, performance history, text messages, camera footage, personnel files, and allegations that could damage reputations. All of that makes casual AI use especially risky. If a manager uploads witness statements, medical information, or personnel documents into an unapproved AI tool, the company may have just created privacy, confidentiality, privilege, and data-security problems. The DOL’s AI best-practices roadmap specifically includes securing and protecting worker data among its recommended approaches for responsible workplace AI.


Discovery is another concern. If AI helped summarize evidence, generate interview questions, assess witness statements, or draft findings, a plaintiff’s lawyer may ask what tool was used, what data was uploaded, what prompts were entered, what outputs were generated, whether the outputs were preserved, and how much the final decision depended on them. That doesn’t mean AI can’t be used. It means the company needs a protocol before anyone starts experimenting.


State and local law is moving faster than many employer policies

Federal law is relevant, but employers shouldn't stop there. New York City’s automated-employment-decision-tool law requires a bias audit, public audit information, and notices before employers and employment agencies use covered automated employment decision tools. The New York City Department of Consumer and Worker Protection says the law applies to automated employment decision tools and identifies required bias audits, public posting, and notice obligations.


California’s Civil Rights Council secured approval for employment regulations addressing automated-decision systems, with rules set to take effect October 1, 2025. California’s Civil Rights Department described the regulations as addressing employment discrimination risks tied to automated-decision systems.


Colorado’s automated-decision framework has also been moving. The Colorado Attorney General explains that Senate Bill 26-189, signed in May 2026, repealed and reenacted the earlier provisions and created new requirements for automated decision-making technology used to materially influence consequential decisions, with the new provisions going into effect January 1, 2027.


Maryland already restricts use of facial recognition services during applicant interviews unless the applicant consents through a specified waiver.


The exact scope of these laws varies. Some apply most clearly to hiring or promotion tools, not every internal investigation. But the trend is still important. Regulators are increasingly focused on whether employers can explain, audit, disclose, and defend the use of automated systems in employment-related decisions.


What Fractional General Counsel and Outside General Counsel should put in place

A business doesn’t need to ban AI from investigations but the process does need rules.

A practical protocol should address:

  • which AI tools may be used;

  • who may approve AI use in an investigation;

  • what information may never be uploaded;

  • whether employee data can be used for model training;

  • how prompts and outputs will be preserved;

  • when legal review is required;

  • whether the tool is being used only to organize information or to influence findings;

  • whether the investigation involves protected activity, medical information, accommodation issues, leave rights, harassment, discrimination, retaliation, or whistleblower allegations;

  • how the company will document human review;

  • how the final report will distinguish facts, AI-assisted summaries, witness statements, and the investigator’s findings.


The company should also train Human Resources, (HR), managers, and investigators not to use personal AI accounts or unapproved tools for investigation work. A well-written AI policy doesn’t help much if the people handling the most sensitive information in the business are pasting it into whatever tool is easiest at the moment.


The final investigation report should not read like software made the decision

A good investigation report should show that the company followed a fair process. That means the report should identify the complaint, the scope of review, the witnesses interviewed, the documents considered, the factual findings, the evidence supporting those findings, and the basis for any conclusion. If AI was used, the company should be able to explain the limited role it played.

AI can help make the report cleaner. It can’t supply judgment the investigator didn’t exercise.

For serious matters, the safer practice is to treat AI output as a draft aid or organizational tool, not as the conclusion. The final findings should come from a trained human decision-maker who reviewed the record, considered the evidence, and can explain the reasoning.


Conclusion

Artificial Intelligence in Workplace Investigations is going to become more common because the practical benefits are real. Investigations often involve too much data, too little time, and too much pressure on HR and management to move quickly. But speed isn’t the only measure of a defensible investigation.


Employers need to know what tool is being used, what data is being uploaded, how the tool works, what the vendor does with the information, whether bias has been evaluated, whether state or local law applies, and whether a human being actually made the findings.

Used carefully, AI can help organize the evidence room. Used carelessly, it can become the reason the investigation gets attacked.


If your business is considering AI tools for HR investigations, employee complaints, internal reviews, or disciplinary decisions, Todd Nurick and Nurick Law Group, LLC can help develop the policies, vendor review, investigation protocols, and broader Fractional General Counsel or Outside General Counsel support needed to reduce avoidable risk.


Sources

Disclaimer: This article is for informational purposes only and isn't legal advice. Reading it doesn't create an attorney-client relationship. Todd Nurick and Nurick Law Group aren't your attorneys unless and until there is a fully executed written fee agreement with Todd Nurick or Nurick Law Group.

 

© 2025 by Nurick Law Group. ***Nurick Law Group and Todd Nurick do not function as your legal counsel or attorney unless a fee agreement has been established. The information presented on this site is not intended to serve as legal advice. Our objective is to educate businesses and individuals regarding legal issues pertinent to Pennsylvania. 

 

bottom of page