Workplace harassment investigations have always relied on a simple assumption: a screenshot, a text thread, or a video either happened or it did not. That assumption is breaking down. Employment attorneys say AI-generated deepfakes, fabricated text messages, and manipulated images are now showing up in real HR investigations, and the standard playbook for confirming what actually happened was not built for evidence that can be faked convincingly in minutes.
The evidence problem AI creates
Tracey Diamond, a partner in the labor and employment practice at Troutman Pepper Locke, has watched the shift firsthand. “It’s sort of like a continuation of the old version of employees lying,” she said. “They’re just creating evidence to further their lies.” What has changed is not the impulse to fabricate, which HR has managed for decades, but the quality of the fabrication. “We’re not seeing it a lot yet, but I do think that the possibilities are becoming more real because AI is getting more and more accurate,” Diamond said.
The risk cuts both ways. A complainant can submit a fabricated recording or image to support a false claim, and an employee accused of real misconduct can just as easily claim that genuine evidence was AI-generated. Either way, the investigator is left holding evidence they can no longer take at face value, in a process, per HRTech Edition’s coverage of automated termination decisions, where getting the facts wrong already carries real legal exposure.
What makes this different from ordinary workplace lying is speed and scale. Fabricating a convincing paper trail used to take effort and left traces: mismatched fonts, awkward phrasing, timestamps that did not line up. Generative tools collapse that effort to minutes and increasingly erase the traces along with it. An investigator who once had to catch a bad forger now has to assume a good one, and the standard of proof that used to feel like due diligence now looks more like the minimum.
The cases already testing the line
Two disputes now working through the courts show how this plays out when it is not hypothetical. A Washington state trooper has sued his employer alleging that a supervisor used AI to create and circulate a deepfake video depicting him kissing a coworker, and is now pursuing claims for discrimination, retaliation, and invasion of privacy. In California, an appellate court upheld a four million dollar verdict for a police captain who was subjected to a hostile work environment after a sexually explicit AI-generated image resembling her circulated among colleagues, finding it met the legal standard for unlawful harassment.
Federal law is catching up in pieces. The Take It Down Act, which took effect in 2025, makes distributing non-consensual intimate images generated or altered by AI a federal crime and requires platforms to remove reported content within 48 hours of notice. The Equal Employment Opportunity Commission has separately stated that circulating sexually demeaning AI-generated images or deepfakes of a coworker is itself a form of harassment that can violate Title VII, meaning employers do not get to treat “the video was fake” as a defense to a real complaint.
What “proper investigatory technique” now means
Diamond’s guidance to HR teams is procedural, not technological. Investigators, she said, “should follow proper investigatory techniques in terms of seeing what they could find to corroborate the story,” rather than accepting a single piece of digital evidence as conclusive. In practice that means cross-checking: “if someone is saying that an employee sent them a harassing text message, checking the employee’s phone as well as the alleged harasser’s phone to see if the messages are on both phones.”
For higher-stakes cases, that corroboration should not stop at the HR team. “If it’s an investigation that involves potential termination or a high-level situation that could put the company at risk, we recommend that they use outside experts,” Diamond said. She also argues employers should say the obvious part out loud in policy. “While you’d think it goes without saying that AI should only be used to tell the truth, maybe it warrants saying that in a written policy.”
What it means for the HR leader
The practical shift is a change in the burden of proof inside an investigation. Where a screenshot or recording used to close a question, it now opens one: can this be corroborated independently, and by whom. HR teams that keep treating digital evidence as self-authenticating are exposed twice over, once to a false complaint built on fabricated material, and once to a real complaint dismissed because a genuine recording looked too convenient to trust. Get the call wrong in either direction and the company is exposed, a dynamic HRTech Edition has tracked as courts write HR AI’s rulebook in real time. Diamond’s read on the legal exposure cuts in the employer’s favor only when the process was sound: “If an employer ultimately gets it wrong, but after they conducted a thorough investigation and it was reasonable for them to reach that conclusion because the technology was that good, then that should protect them.”
What to do now
Three moves are within reach before the next complaint lands. First, write an explicit AI-use policy that states evidence must be truthful and outlines consequences for fabricating it, so HR is not relying on an unstated assumption. Second, build a standing corroboration step into every investigation involving digital evidence, meaning at least one independent source is checked before a finding is reached. Third, set a threshold, such as any case involving possible termination, above which outside forensic or legal expertise is brought in rather than left to internal judgment alone. None of this requires HR to become a digital forensics unit. It requires treating a single unverified file the way a careful investigator already treats a single uncorroborated witness. The employers who adapt fastest will be the ones who write the policy before a fabricated recording forces their hand, not after.
Source: Troutman Pepper Locke
