Emotion AI Is Quietly Infiltrating Workplaces

Emotion AI — software that claims to detect and interpret human feelings through facial expressions, voice tone, and body language — is quietly embedding itself into workplaces across the United States and beyond. A sweeping new feature report from The Atlantic, written by Ellen Cushing, reveals how these scientifically dubious tools are becoming a routine part of hiring, performance reviews, and daily work interactions, often without employees fully understanding what is happening.

The report raises urgent questions about worker privacy, scientific validity, and the unchecked proliferation of AI systems that lack a credible evidence base.

Key Takeaways From the Atlantic Report

• Emotion AI software is being deployed in hiring processes, customer service monitoring, and employee performance evaluation across multiple industries
• The underlying science — that facial expressions reliably correspond to specific internal emotions — has been widely debunked by psychologists
• Workers are often unaware that their emotional states are being algorithmically assessed during routine interactions
• The global affective computing market is projected to exceed $140 billion by 2032, according to some industry estimates
• Regulatory frameworks in the U.S. remain largely absent, though the EU's AI Act has begun to address emotion recognition systems
• Bias and accuracy concerns are significant, with studies showing these tools perform differently across racial, gender, and cultural groups

What Emotion AI Actually Does — and Doesn't Do

Affective computing, the broader field that encompasses emotion AI, has roots going back to MIT researcher Rosalind Picard's work in the late 1990s. The basic premise is that AI systems can analyze micro-expressions, vocal patterns, eye movements, and other physiological signals to infer a person's emotional state.

Companies like HireVue, Affectiva (now part of Smart Eye), and Realeyes have built commercial products around this concept. HireVue, for instance, previously used facial analysis in its video interview screening tool before dropping that feature in 2021 amid public backlash. However, many lesser-known vendors continue to offer similar capabilities with far less scrutiny.

The fundamental problem, according to a landmark 2019 meta-analysis by psychologist Lisa Feldman Barrett and colleagues, is that the core assumption is wrong. Their review of over 1,000 studies found that facial movements do not reliably map onto specific emotional states. A furrowed brow does not necessarily mean anger. A smile does not necessarily mean happiness. Context, culture, and individual variation make these inferences deeply unreliable.

Workplace Deployments Are Growing Despite Scientific Criticism

Despite the scientific community's pushback, workplace adoption of emotion AI continues to accelerate. The Atlantic report details how these tools now appear in call centers, where supervisors receive real-time dashboards showing agents' supposed emotional states. They show up in corporate training platforms, where employees' 'engagement levels' are scored algorithmically.

Some companies use emotion detection during onboarding to gauge new hires' enthusiasm. Others deploy it in virtual meetings to assess team morale. The common thread is that employers are making consequential decisions — about promotions, terminations, and performance ratings — based on outputs from systems that lack scientific credibility.

This is not a fringe phenomenon. Market research firm MarketsandMarkets has valued the emotion detection and recognition market at approximately $23.5 billion in 2022, with compound annual growth rates exceeding 12%. Unlike the large language model boom driven by companies like OpenAI, Anthropic, and Google DeepMind, the emotion AI sector operates with far less public visibility and far fewer guardrails.

The Bias Problem Compounds the Science Problem

Even if the underlying science were sound — which experts broadly agree it is not — emotion AI tools carry significant algorithmic bias risks. Multiple studies have shown that these systems interpret Black faces as angrier than white faces, even when displaying identical expressions. Women's expressions are more frequently coded as emotional compared to men's.

The implications for workplace equity are severe:

• A job candidate could be rejected because the AI misread cultural differences in facial expression
• A call center worker could receive negative performance marks because the system misinterpreted their neutral expression as disengaged
• Employees with disabilities affecting facial movement or vocal tone could be systematically disadvantaged
• Workers from non-Western cultural backgrounds, where emotional expression norms differ, face heightened misclassification risks

These are not hypothetical scenarios. They represent the logical consequences of deploying pattern-matching systems trained on narrow, often Western-centric datasets in diverse workplace environments.

Regulation Lags Behind Deployment

The regulatory landscape for emotion AI remains fragmented and largely toothless in the United States. Illinois' Biometric Information Privacy Act (BIPA) offers some protections, and New York City's Local Law 144 requires audits of automated employment decision tools. But neither law specifically addresses the pseudoscientific foundations of emotion recognition technology.

The European Union's AI Act, finalized in 2024, takes a stronger stance. It explicitly restricts the use of emotion recognition systems in workplaces and educational settings, classifying them as high-risk applications. This creates a significant transatlantic gap — European workers gain meaningful protections while American workers remain largely exposed.

Compare this to the regulatory response around large language models, where governments have at least engaged in high-profile debates about safety and transparency. Emotion AI has flown under the radar, partly because its vendors are smaller and less visible than companies like OpenAI or Meta, and partly because the technology integrates quietly into existing HR and management software suites.

Why This Matters for the Broader AI Industry

The spread of emotion AI in workplaces is a cautionary tale for the entire artificial intelligence sector. It demonstrates what happens when commercial incentives outpace scientific validation and regulatory oversight.

Several broader lessons emerge:

• Market demand alone does not validate a technology's scientific basis
• Enterprise buyers often lack the technical expertise to evaluate AI vendors' claims critically
• The 'AI' label can lend undeserved credibility to products built on shaky foundations
• Worker consent and transparency mechanisms remain woefully inadequate
• The absence of regulation creates a vacuum that commercial actors will fill

The emotion AI phenomenon also highlights a troubling dynamic in enterprise software procurement. Decision-makers — typically HR leaders and operations executives — are purchasing these tools based on vendor marketing materials rather than peer-reviewed evidence. The result is a growing ecosystem of workplace surveillance technology dressed up in the language of employee wellbeing and organizational health.

What Workers and Companies Should Know

For individual workers, awareness is the first line of defense. Employees should ask their employers directly whether any form of emotion detection or sentiment analysis is being used in evaluations, meetings, or hiring processes. In jurisdictions with relevant privacy laws, workers may have legal grounds to refuse participation.

For companies considering these tools, the calculus should be straightforward. The scientific consensus is clear: current emotion AI technology cannot reliably do what it claims to do. Deploying it exposes organizations to legal liability, reputational damage, and the very real risk of making worse decisions than they would without it.

HR technology vendors, meanwhile, face a reckoning. As regulatory pressure mounts — particularly from Europe — and as public awareness grows through reporting like The Atlantic's feature, the market for pseudoscientific emotion detection may contract. Companies that have built their business models around these capabilities will need to pivot or face obsolescence.

Looking Ahead: The Future of Emotion AI Regulation and Research

The next 2 to 3 years will likely prove decisive for emotion AI in the workplace. The EU AI Act's provisions will begin taking effect, potentially creating a de facto global standard as multinational companies harmonize their practices across regions.

In the U.S., state-level action may accelerate. California, Washington, and Massachusetts are all considering legislation that could restrict automated emotion recognition. Federal action remains unlikely in the near term, but the FTC has signaled interest in scrutinizing AI products that make deceptive claims about their capabilities.

On the research front, some scientists are exploring more nuanced approaches to affective computing that account for context, individual baselines, and cultural variation. These approaches may eventually yield more defensible tools. But they are years away from commercial readiness, and they will require fundamentally different architectures than the products currently being sold.

Until then, the uncomfortable reality remains: millions of workers are being evaluated by AI systems that cannot do what their makers claim. The Atlantic's report serves as a critical reminder that not all AI is created equal — and that the label 'artificial intelligence' should never be a substitute for scientific rigor.