Your AI Meeting Tool's Sentiment Scoring Is Already Illegal in Europe — And Fines Start at €35 Million

Some of the most popular AI meeting tools on the market today don't just transcribe what you say — they claim to measure how you feel while you say it. Engagement scores. Sentiment graphs. Participant attention ratings. Features marketed as "meeting analytics" that promise to tell managers which employees were focused and which "checked out."

There's just one problem: in the European Union, using AI to infer the emotions of workers is now explicitly illegal.

Article 5(1)(f) of the EU AI Act prohibits "the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions." This isn't a future regulation. The prohibition on workplace emotion recognition took effect on February 2, 2025, and deploying a prohibited AI practice carries fines of up to €35 million or 7% of global annual turnover — whichever is higher.

Yet a stunning number of AI meeting tool vendors are still selling these features. Still demoing them at conferences. Still writing them into enterprise contracts. As the August 2, 2026 deadline approaches for the full enforcement of high-risk AI system requirements, any organization using these tools with European employees is sitting on a compliance time bomb.

What the EU AI Act Actually Bans

The EU AI Act is the world's first comprehensive AI regulation, and its prohibited practices list is not subtle. The Act bans AI systems that manipulate behavior, exploit vulnerabilities, perform social scoring — and critically, infer emotions in the workplace from biometric data.

The European Commission's official guidance makes clear that the workplace context should be interpreted broadly, covering any physical or virtual space where work is performed. That includes every Zoom call, Microsoft Teams meeting, and Google Meet session where an AI tool is analyzing participants' emotional states.

According to the European Commission's regulatory framework, the prohibition entered into force in February 2025, and the full high-risk AI system requirements become enforceable on August 2, 2026. The penalties are among the steepest in global tech regulation.

Which Meeting Tools Are Affected?

The tools most squarely in the crosshairs are those offering what the industry euphemistically calls "meeting analytics" — features that go beyond transcription into emotional inference and behavioral scoring of participants.

Consider tools like Read.AI, which openly describes its functionality: the platform analyzes "facial and verbal cues from attendees to determine whether they are expressing positive, neutral or negative reactions to the meeting in real time." Individual participants receive engagement and sentiment scores, creating a per-person behavioral profile of every meeting. Read.AI has acknowledged the regulatory risk for its EU users by noting that "for users in the EU and UK, the Read Score does not include any use of meeting video or facial elements."

But the ban's scope is wider than facial analysis alone. The Future of Privacy Forum's analysis of the prohibition notes that it covers emotion recognition from biometric data, which under the AI Act's definition includes voice characteristics like pitch, volume, and intonation — the exact signals many AI meeting tools analyze for "sentiment" scoring.

Other platforms in the broader conversation intelligence space — including tools like Gong, Chorus.ai (ZoomInfo), and various contact center analytics platforms — offer sentiment analysis features that detect emotional shifts during calls through vocal pattern analysis. When these tools are deployed to analyze employee meetings and performance, they fall directly within the prohibited category.

Why "Sentiment Analysis" Is Not Just "Transcription"

There's a critical legal distinction here. As UC Today's analysis of the ban explains, the EU did not ban AI in the workplace. Transcription is fine. AI-generated summaries are fine. Action items, decisions captured, and searchable meeting archives are all fine. What's banned is specifically "the inference of a person's internal emotional state from their biometric data."

That means a tool that converts speech to text is compliant. A tool that converts speech to text and then scores whether the speaker sounded enthusiastic, frustrated, or disengaged — that's the one that's illegal when applied to employees in the EU.

The Employer Liability Problem

The regulatory burden doesn't fall only on the AI vendors. Under the EU AI Act's framework, both providers (companies that build AI systems) and deployers (organizations that use them in a professional capacity) bear compliance obligations. If your company purchases an AI meeting tool and enables its emotion recognition features for meetings involving EU-based employees, your organization is a deployer of a prohibited AI system.

This mirrors the employer liability dynamic already emerging in the United States around AI meeting tools. As employment attorneys at Littler Mendelson warned in a February 2026 analysis of the In re Otter.AI Privacy Litigation, even organizations that did not build these tools are responsible for how they are deployed in the workplace. The HR Executive coverage of the Otter.ai lawsuit highlighted that "the AI transcription and recording issue is a hot issue" — and that HR leaders should act now before courts define the boundaries for them.

For multinational employers, the compliance picture is especially complex. A single virtual meeting connecting employees in New York, London, and Berlin may simultaneously trigger U.S. state wiretapping laws, GDPR consent requirements, and the EU AI Act's emotion recognition prohibition. As explored in our article on Otter.ai's federal hearing and the EU AI Act deadline, the regulatory walls are closing in from multiple directions simultaneously.

The GDPR Multiplier Effect

The EU AI Act doesn't operate in isolation. The IAPP's mapping of the interplay between the EU AI Act and GDPR shows that the two regulatory frameworks overlap significantly. Voice recordings in meetings constitute personal data under GDPR. Emotional inference from those recordings triggers additional processing obligations.

Under GDPR Article 5, processing must adhere to data minimization — collecting only what's necessary for a defined purpose. If your stated purpose is meeting transcription, analyzing participants' emotional states goes well beyond that scope. Meanwhile, Article 9 places special restrictions on processing biometric data, requiring explicit consent that — as multiple legal analyses have noted — is nearly impossible to obtain freely in an employer-employee relationship due to the inherent power imbalance.

The practical consequence: organizations using emotion-inferring AI meeting tools in Europe face potential liability under both the AI Act and GDPR simultaneously. As one legal analysis noted, combined fines could reach up to 11% of global turnover.

Why Your Vendor Hasn't Told You

If workplace emotion recognition has been illegal in the EU since February 2025, why are so many vendors still selling these features? The uncomfortable answer, according to industry observers, is a combination of market pressure, geographic ambiguity, and a calculated gamble on enforcement timing.

Many vendors have global customer bases and have been slow to segment their feature sets by jurisdiction. Others are banking on the fact that enforcement infrastructure is still being built — national regulators are only now setting up their AI Act oversight mechanisms, and the first major enforcement case is widely expected this year.

But as Social Europe's analysis of AI note-takers at work observed, these tools transmit data to U.S.-based servers, creating cross-border transfer issues that compound the emotion recognition ban. A tool that infers emotions and sends that data outside the EU without proper safeguards is stacking violations.

The regulatory patience is running out. The European Commission's November 2025 review of the AI Act specifically declined to soften the prohibited practices list. The bans are staying. And with the August 2, 2026 high-risk enforcement deadline now less than 75 days away, the window for quiet non-compliance is slamming shut.

What Compliant Meeting AI Actually Looks Like

The good news is that the EU AI Act draws a precise line, and meeting tools that respect it can continue operating without issue. The regulation targets one specific behavior: inferring internal emotional states from biometric data in the workplace. Everything else — transcription, summarization, action item extraction, speaker identification, searchable archives — remains fully legal.

But the safest architecture goes further. As we explored in our analysis of AI meeting bots and the chilling effect on workplace speech, the fundamental privacy risk with cloud-based meeting tools isn't just what features they offer today — it's the data they collect that could be analyzed tomorrow. When your voice data sits on a third-party server, any future feature addition or policy change can retroactively expose it.

Apple's approach to AI reflects this same philosophy. As AppleInsider reported just yesterday, Apple's 2026 AI strategy continues to prioritize on-device processing through Apple Foundation Models running on the Neural Engine, with Private Cloud Compute as a fallback that still ensures "data is not stored or made accessible to Apple." This architectural commitment to privacy aligns perfectly with the EU AI Act's regulatory direction.

Basil AI was built on this exact principle. Every recording is transcribed entirely on your device using Apple's on-device Speech Recognition framework. No audio or transcript data ever touches a cloud server. There are no engagement scores, no sentiment graphs, no behavioral analysis — because those features require the kind of data collection that regulatory frameworks worldwide are now prohibiting.

What you get instead is what actually matters for meeting productivity: accurate transcription, smart summaries, action items, and seamless export to Apple Notes — all processed locally, all under your complete control, all fully compliant with every jurisdiction on the planet.

What You Should Do This Week

If your organization uses AI meeting tools with any employees or partners based in the EU, here's the immediate action plan:

1. Audit your AI meeting tools — Check whether any tool you've deployed offers sentiment analysis, engagement scoring, emotion detection, or attention monitoring. If it does, determine whether those features are active for EU-based users.
2. Check your vendor's EU compliance posture — Ask explicitly: "Does your tool infer emotions from biometric data? If so, have you disabled this feature for EU deployments?" Get the answer in writing.
3. Review your Data Processing Agreements — Ensure your DPAs with AI meeting tool vendors account for AI Act obligations, not just GDPR. The two frameworks have overlapping but distinct requirements.
4. Disable emotion-inferring features immediately — If you're using a tool with sentiment scoring in EU meetings, turn those features off now. The prohibition has been active since February 2025.
5. Consider on-device alternatives — For the highest compliance certainty, switch to tools that process audio entirely on-device, eliminating the data collection that makes emotion recognition possible in the first place.

The EU AI Act represents a fundamental shift in how workplace AI is regulated. For AI meeting tools specifically, the message is clear: transcribe what people say, help them act on it, but don't presume to measure what they feel. The tools that respect that line will thrive. The ones that don't will face a €35 million reckoning.

Meeting Notes Without Surveillance

Basil AI transcribes your meetings 100% on-device. No cloud processing, no sentiment scoring, no behavioral analysis. Just accurate notes that stay completely private — and fully compliant with every regulation worldwide.