AI tools that monitor worker emotions are growing, and so are the questions around their use
Picture this: an AI tool scans your employee’s face during a video call and decides they seem disengaged. Another listens to a customer service rep’s tone and flags it as tense. A third monitors your team’s Slack messages and surfaces a report telling you which department seems most unhappy this week.
This isn’t science fiction. It’s happening now, in workplaces across the country, and most HR leaders haven’t had a conversation about it yet.
The technology is called emotion AI, or affective computing, and it encompasses tools that analyze facial expressions, voice tone, word choice, and biometric signals to infer how workers are feeling. And while it’s most prevalent in call centers, trucking, and the fast-food service industry, it’s now expanding into office environments.
“AI is just sort of a new tool for an existing practice,” said Ani Huang, President of Policy and Practice at the CHRO Association in Washington, D.C., which represents the chief human resources officers of major multinational corporations. Employers have monitored calls, emails, keystrokes, and camera feeds for years. Emotion AI, she suggests, simply extends that logic at a scale no human manager ever could.
That may be true. But the science behind it is contested, the law hasn’t caught up, and the decisions HR leaders make now will have lasting consequences for employee trust. Here is where things stand.
Is it even accurate?
Before any governance or legal conversation can happen, there is a more fundamental question: do these tools actually work?
Jennifer Rubin, a partner at Mintz in San Diego, California who specializes in employment law, is skeptical. When she first encountered emotion AI tools, her instinct was not legal alarm. It was disbelief.
“If humans can’t pick up on these cues in an in-person interaction, how can a machine possibly pick up on these cues using a camera?” she said. “I’m going to use the term troubling, because I really wonder about the accuracy.”
Rubin drew a parallel to pre-employment assessments and workplace evaluations, which must be carefully calibrated and comply with state and federal anti-discrimination law. No equivalent requirements yet govern emotion AI tools.
“How are these systems being calibrated? Why do we think they’re reliable?” she asked.
The concern goes beyond reliability. People express emotion differently depending on culture, context, personality, and circumstance. A scowl of concentration can read as anger. A flat affect can register as disengagement.
Joseph Lazzarotti, a principal at Jackson Lewis and leader of the firm’s Privacy, AI and Cybersecurity practice group in Tampa, Florida, pointed to the discrimination risk this creates.
“There’s potentially opportunities for disparate impact in terms of how certain people are categorized or how the emotions are ranked,” he said. “It can get pretty interesting once you start having a tool that can rank people lower based upon the perception of their emotion during the time of an interview.”
READ MORE: AI adoption raises psychosocial risks at work, ILO warns
Rubin put it more plainly.
“At the end of the day, how can a machine know me? My family, my friends, my spouse know me. But how would a machine know me?”
A legal grey area
There is no federal law in the United States specifically governing emotion AI in the workplace. What exists is a patchwork of state-level rules that vary considerably and are still being written.
Lazzarotti pointed to New York, New York City, and California as jurisdictions with relevant statutes or regulations, with Colorado expected to introduce new requirements in 2027. For multi-state employers, this creates a compliance puzzle that is already complex and growing more so.
One area drawing particular attention is biometrics. Lazzarotti cited an ongoing Illinois case involving Fireflies, a transcription tool, in which plaintiffs allege the software captures a voice print to assign speech to the correct speaker. Under the Illinois Biometric Information Privacy Act, a voice print is biometric data, requiring employee consent, a written policy, and significant financial penalties if those steps are skipped. Emotion AI tools that analyze vocal tone or facial expressions may face the same classification.
“Some of the emotions that can be captured might be captured by facial expressions, might be captured by voice intonation, and that could constitute a biometric in some cases,” Lazzarotti said.
Rubin added that employers must also consider what happens to the data once it’s collected. Depending on how emotion AI output is stored and used, it could potentially be classified as part of an employee’s health record, triggering a separate set of strict obligations.
The European Union has taken the strongest position yet, banning workplace emotion AI outright under Article 5 of the EU AI Act, citing limited reliability, discrimination risk, and the inherent power imbalance in the employment relationship, with narrow exemptions for medical and safety purposes. For U.S.-based companies with European operations, the rules may already apply depending on how and where the tools are deployed. Back home, a federal equivalent is nowhere in sight.
“If I had to bet, I would say not anytime soon,” Lazzarotti said. “Congress has tried for almost two decades to get something at the federal level on data privacy and has not been successful. Even if they believe strongly in the need for regulation around AI, some of those issues may prevent them from reaching agreement.”
Handle with care
Amid the reliability and legal concerns, Huang offered a perspective that tends to get lost: used carefully, emotion AI could genuinely help employees.
“HR has always wanted to be able to have that relationship or connection with every employee at the company,” she said. “And it’s just never been possible. The hope behind this is that it might actually identify, in a lot of cases, that somebody needs help before it’s gotten to a really bad place.”
Huang drew an important distinction between AI that surfaces aggregate signals, such as tension within a team, and AI that reports on individual employees and informs employment decisions. The former may have real value. The latter is where the legal and ethical exposure lies.
READ MORE: AI is improving performance measurement. Pay is still human
Transparency, she said, is essential, which means being specific with employees about how the data is being used and what role it plays in decisions about them.
“The AI is not making any type of decision. It’s not doing your performance review based on this,” she said. “This is a data point that we think is going to help us help you versus something we’re using against you. I think that’s the most important thing.”
Rubin’s advice to any employer considering these tools begins with a single question: why?
“What are the business goals that you believe will be accomplished by putting these devices into play and are there other mechanisms where you can accomplish this?” she said. “I always start with a real deep dive into the why, because I think asking that question can yield lots of interesting information.”
Lazzarotti noted that the human dimension may ultimately be the hardest challenge of all
“Maybe the biggest challenge of all isn’t the union, the statutes, or the vendor,” he said. “It’s trying to deal with the employee relations issues that come with something like this. People start thinking of “1984.” And so, do you really want that information?”
The technology is still developing. The regulation is still catching up. But the decisions HR leaders make now, about governance, transparency, and purpose, will define how this plays out inside their organizations.
