Unadhi Guides

Emotion AI at work and with customers: an EU AI Act check sheet (Article 5(1)(f) ban, Article 50(3) notice) for call centres, shops and hiring

Updated 27 Oct 2026 · Disclosure: the EURO AI Register mentioned below is an Unadhi product with a free edition. This guide is for information only, not legal advice.

What you get: a finished emotion AI check sheet in three tabs: a feature list where you enter every AI feature that scores mood, tone, stress, "sentiment" or engagement (call analytics, interview tools, meeting assistants, in-store cameras, learning apps), with a result formula that sorts each one into STOP, REVIEW, ALLOWED WITH DUTIES or OUT OF SCOPE; a notice record for the Article 50(3) duty to tell people; and a dates and actions tab, plus five made-up example rows.

Short answer: two questions decide almost everything. Whose emotions? Using AI "to infer emotions of a natural person in the areas of workplace and education institutions" has been banned since 2 February 2025, except for medical or safety reasons (Article 5(1)(f)). From what input? The emotion recognition rules only cover inferences "on the basis of their biometric data" such as voice, face or typing rhythm (Article 3(39)); the Commission says inferring emotions from written text is out of scope (Guidelines C(2025) 5052, paras. 251, 265). Emotion recognition on customers is not banned by Article 5(1)(f), but since 2 August 2026 deployers must "inform the natural persons exposed thereto" (Article 50(3)), and it is listed as high-risk in Annex III, point 1(c).

Why this matters now

Many customer service, sales call and video meeting tools now ship a "sentiment", "mood" or "engagement" score switched on by default. If the score is about your own staff, job candidates or students, and it is built from their voice, face or keystrokes, it can fall under the ban, with fines for breaching Article 5 of up to EUR 35 million or 7% of worldwide annual turnover, and for SMEs the lower of the two (Article 99(3) and (6)). The Commission's own examples include "Using webcams and voice recognition systems by a call centre to track their employee's emotions, such as anger, is prohibited" and "Using emotion recognition AI systems during the recruitment process is prohibited" (Guidelines, para. 254).

Step 1: tab 1, feature list (one row per AI feature)

ColumnWhat to writeSource
A. Tool and featureFor example "Call software: agent sentiment score" or "Hiring tool: video interview tone analysis". Check settings pages for words like sentiment, mood, tone, stress, engagement, frustration
B. Whose emotionsStaff (employees, contractors, trainees, volunteers, job candidates), Student (learners of an education or training institution, including applicants), or Customer (customers, visitors, callers)Guidelines paras. 254, 255 (workplace read broadly, includes candidates and shops)
C. InputText only (chat, email, reviews), or Biometric (voice, face or body video, keystrokes, eye tracking, heartbeat or other wearable data)Art. 3(34), 3(39); Guidelines paras. 251, 265
D. OutputEmotion or intention (angry, happy, stressed, bored, satisfied, about to buy), Physical state (fatigue, pain), or Expression only (counts smiles or raised voices without concluding anything about feelings)Recital 18; Guidelines paras. 247 to 249
E. PurposeMedical or safety (protecting life and health, for example near dangerous machines; therapeutic use of a CE-marked medical device), Personal training only (results not shared with HR and no effect on assessment), or Other (quality, sales, wellbeing, productivity)Guidelines paras. 254, 256 to 258, 263
F. Result=IF(C2="Text only","OUT OF SCOPE of emotion recognition rules: GDPR still applies",IF(D2<>"Emotion or intention","NOT EMOTION RECOGNITION: keep a note why",IF(B2="Customer","ALLOWED WITH DUTIES: tell people (tab 2), GDPR, high-risk rules from 2 Dec 2027",IF(E2="Other","STOP: banned since 2 Feb 2025, switch it off","REVIEW: narrow exception, get a written expert view"))))Formula and labels are our working rule, not a legal test

Read the narrow exceptions carefully. The Commission says the medical or safety exception "should be narrowly interpreted", that "general monitoring of stress levels at the workplace is not permitted", that a system to detect burnout or depression at work "would remain prohibited", and that safety means protecting life and health, not property "against theft or fraud" (Guidelines paras. 256 to 258). The training-only route is allowed only "if the results are not shared with HR responsible persons and cannot impact the assessment, promotion etc." of the person trained (para. 254).

Step 2: tab 2, notice record (every ALLOWED WITH DUTIES row)

ColumnWhat to writeSource
A. FeatureCopied from tab 1
B. Where people are toldFor example the call greeting ("This call is analysed by AI to detect tone of voice"), a sign at the shop entrance, the video call inviteArt. 50(3): inform the persons exposed "of the operation of the system"
C. WhenMust be before or at the first interaction or exposure. Write the exact momentArt. 50(5): "at the latest at the time of the first interaction or exposure"
D. Clear and accessible?Yes or No: spoken and written versions, readable font, not hidden in termsArt. 50(5): "clear and distinguishable", "applicable accessibility requirements"
E. Privacy notice updated?Yes or No, with a link to the section that covers voice or video analysisArt. 50(3): process personal data in accordance with Regulation (EU) 2016/679
F. Check=IF(OR(B2="",C2="",D2<>"Yes",E2<>"Yes"),"FIX before next use","OK")Our working rule

Step 3: tab 3, dates and actions

DateWhat appliesYour actionSource
2 February 2025Article 5 bans, including 5(1)(f) emotion inference at work and in educationEvery STOP row: switch the feature off and record the dateArt. 113(a)
2 August 2026Article 50 transparency duties, including 50(3) for emotion recognitionEvery ALLOWED WITH DUTIES row: tab 2 completeArt. 113 (general date)
2 December 2027High-risk rules for Annex III systems, which include "AI systems intended to be used for emotion recognition" (Annex III, 1(c))Ask the vendor how it will meet the high-risk requirements; list your deployer dutiesArt. 113(c)(i) as amended by the Digital Omnibus

Customer-facing emotion AI can still be banned in other ways: the Commission notes such systems "may be prohibited in certain cases by virtue of Article 5(1)(a) and (b)" (harmful manipulation and exploitation), and that data protection and consumer protection law continue to apply (Guidelines para. 266). Member States and collective agreements may also set rules more favourable to workers on AI at work (Art. 2(11), as quoted in Guidelines para. 264).

Five made-up example rows

FeatureResultWhat happened
Small call centre: call software scores each agent's voice for "frustration" and sends a weekly chart to the team leadSTOPAgent scoring switched off in settings; date recorded
Same software: flags calls where the customer's voice sounds angry so a senior agent can joinALLOWED WITH DUTIESGreeting now says calls are analysed by AI for tone of voice; privacy notice updated; vendor asked about the high-risk rules
Web shop: AI tags support emails and reviews as positive, neutral or negativeOUT OF SCOPE of emotion recognition rulesText only; kept in the GDPR records of processing
Recruiting agency: video interview tool rates candidates' "confidence" and "enthusiasm" from face and voiceSTOPFeature removed from the hiring process; candidates assessed by people
Delivery firm: dashcam AI warns drivers when it detects signs of fatigueNOT EMOTION RECOGNITIONFatigue is a physical state (Recital 18); note kept, GDPR checks still done

These rows are made up for illustration and do not describe any real business.

This sheet is a working aid, not a legal opinion or proof of compliance. References are to Regulation (EU) 2024/1689 (Articles 3(34), 3(39), 5(1)(f), 50(3) and 50(5), 99(3) and 99(6), 113, Annex III point 1(c), Recitals 18 and 44, as shown on artificialintelligenceact.eu with the Digital Omnibus amendments) and to the Commission Guidelines on prohibited artificial intelligence practices, C(2025) 5052 final of 29 July 2025, section 7 (paras. 239 to 269), as published on the AI Act Service Desk. The Guidelines are not binding; the Court of Justice gives the final interpretation. The result labels, formulas and the order of the questions are our own working rules. Checked on 27 October 2026. Spot an error in this guide? Write to mininglife7@gmail.com.

For accountants and tax advisers

Clients with call teams, shops or hiring pipelines often do not know that a "sentiment" switch in their software can be a banned practice. Tab 1 takes about fifteen minutes per client and shows which features to switch off and which need a notice. Stay inside your professional rules and refer legal questions on.

Keep the result next to your AI tool list

The EURO AI Register free edition runs in your browser with no account and no upload. It covers your first 10 AI systems and 10 evidence records, the screening questions, duties by role with dates, and a printable Article 4 AI-literacy record. List each tool once, then attach this sheet as the evidence record for its emotion features.

When you need more than 10 systems, the EURO AI Register membership starts at $19 a month and adds unlimited systems, rule updates and email help. Neither tool is legal advice or proof of compliance.

Partner programme opening soon

Partner programme opening soon: join the free waitlist. Planned terms are 50% of each referred sale (75% once you have 5+ paying customers), single level, paid through Gumroad only from revenue received.

Free start: list your AI tools in the EURO AI Register free edition, then add this sheet. Related: Article 5 prohibited practices · Article 50 chatbot disclosure · AI at work notice for staff and applicants · DPIA screening

Sources

← All guides

Unadhi Guides are free, independent explainers. We correct errors as soon as we find them: write to mininglife7@gmail.com.