Pseudoscientific emotion AI is invading the workplace, an Atlantic report shows

The buried issue here isn't just bad science: it's that 'emotion AI' vendors are selling certainty to HR and compliance buyers who have neither the background nor the incentive to demand peer review. The product works commercially precisely because the buyers aren't the ones being measured.

This connects directly to the consent and transparency problems we've been tracking. The Microsoft VS Code story from early May showed how AI instrumentation can be embedded into workflows without meaningful user awareness, and emotion AI in the workplace is the same dynamic at higher stakes. Where the Copilot case involved metadata, this involves inferences about psychological states being logged and acted on. The ethical divergence benchmark piece from May 3 is also relevant: if frontier models encode different value systems with no standardized accountability, layering contested emotion inferences on top of those models compounds the credibility problem rather than containing it.

Watch whether the EU AI Act's high-risk classification for biometric categorization systems produces a first enforcement action against an emotion-detection vendor by end of 2026. A single fine would do more to reshape this market than any number of critical press reports.