Europe Banned Workplace Emotion AI. The US Is Letting It Scale.
The Atlantic published a feature this month on workplace emotion AI that maps where the technology is now used inside American companies. MetLife runs Cogito on its call centers to analyze agents’ pitch and tone. Burger King is piloting a headset chatbot called Patty in 500 restaurants that generates location-level friendliness scores from keywords like “please” and “thank you.” The company calls it a coaching tool. Framery has tested office chairs with biosensors that read heart rate, breathing, and nervousness. Aware integrates with Slack to run sentiment analysis on employee chat. Microsoft Azure offers similar capability through cloud services. MorphCast extends into video meetings, including a Zoom plugin that reads attention, positivity, and excitement off participants’ faces. The global emotion AI market sits around $3 billion today and is projected to reach $9 billion by 2030.
Europe has decided this category is not permitted at work. Article 5(1)(f) of the EU AI Act prohibits AI systems that infer emotions of a person in the workplace and in education institutions. Fines reach €35 million or 7% of global annual turnover. The exception is narrow. Medical or safety purposes can qualify if they pass a proportionality test. A construction site fatigue sensor may. A call center monitor scoring agents on emotional engagement won’t easily fit. The European Commission’s November 2025 Digital Omnibus simplification adjusted a wide range of AI Act obligations but did not touch the prohibited practices. Workplace emotion AI stayed on the list.
The reasoning the regulation puts on paper is a labor-rights argument, not a paternalism argument. The European Commission guidelines cite power imbalances in those environments as what makes emotion inference at work different from emotion inference at a movie theater. Workers cannot meaningfully consent to monitoring the employer controls and benefits from. The prohibition addresses the coercion the consent framework cannot.
The science the products rest on is contested in ways the marketing rarely acknowledges. A 2019 review in Psychological Science in the Public Interest surveyed the evidence behind the assumption that emotional state can be reliably inferred from facial movements. Researchers concluded that facial configurations lack consistency and specificity for emotion categories. Feldman Barrett told The Atlantic that in the US, people scowl when angry only about 35% of the time. The Ekman six-basic-emotions framework MorphCast leans on has been contested in affective science for over a decade. A study by Lauren Rhue found that emotion-recognition AI scored Black NBA players as angrier than their white teammates, in some cases when those players were smiling. The market built on this technology has continued to scale.
The United States has not drawn the categorical line Europe drew. Its rules cover adjacent harms instead. Illinois prohibits employer use of AI that discriminates against protected classes and requires consent for video-interview AI analysis. Maryland regulates facial recognition during interviews. NYC Local Law 144 requires bias audits for automated employment decision tools. California has two pending bills, AB 1883 and AB 1898, on workplace surveillance and workplace AI. Colorado’s AI Act takes effect June 30. None treats workplace emotion AI itself as a prohibited category. The 2026 federal executive order on AI pushes against some state-level AI rules, Illinois included, instead of toward a national workplace-emotion-AI ban.
The Atlantic reported that the EU rule prompted MorphCast to relocate from Florence to the Bay Area. The corporate record is messier than that framing. The parent company has had a US presence since 2017, and the Florence office still exists. The broader regulatory point still holds. A product category one jurisdiction classified as too coercive for the workplace continues to find a permissive home in another.
The trust and safety question for an AI company deploying emotion AI now turns on what jurisdiction the product sits inside in 18 months, what disclosure regime applies, which workers can challenge a performance decision shaped by an emotion score, and what defense the company mounts if a bias finding lands.
A score for friendliness, engagement, or attention demands the performance of an emotional state, whatever the marketing says about measuring one. The tool reads the worker. The worker, given enough time, reads the tool back, and adjusts. Whatever the tool measures on the way out is something different from what was there on the way in. Europe answered by drawing a line. The American version is being negotiated company by company, in environments where the worker has the least leverage to say no.
The line is somewhere. Europe drew it at the workplace door. The American line is being drawn one Slack channel and one call-center headset at a time.