AI Disclosure in Talent Assessment: What Article 50 Requires Now, and What’s Coming in 2027

On August 2, 2026, the EU AI Act's transparency rules took effect. Among other things, they require that people be told when they are interacting with an AI system, and national authorities can enforce it.

Most coverage of the EU’s summer amendments focused on what they delayed. While certain EU AI Act obligations have later implementation deadlines, Article 50’s transparency requirements are already applicable. Since August 2, organizations using AI with people in the EU have been required to say so.

For organizations using conversational AI in assessments and learning experiences, this makes clear and timely disclosure an important part of the user experience. The law sets a minimum for disclosure. For people being assessed by AI, disclosure is also part of the experience, and it shapes whether they trust the results.

What Article 50 requires

Article 50 does not limit how AI can be used. It requires that certain uses be made visible to the people involved. Its duties fall on two parties: providers, who build AI systems, and deployers, who use them. When an organization buys an assessment platform, the vendor is typically the provider and the organization the deployer.

Article 50 sets out four core duties, as summarized in the European Commission's guidelines on transparency obligations:

• Tell people when they are interacting with AI. Providers must design systems so people are clearly informed when they interact with an AI system, unless that is obvious to a reasonably well-informed person.
• Watermark AI-generated content. Providers of systems that generate text, images, audio, or video must mark outputs in a machine-readable format so they can be detected as AI-generated.
• Inform people exposed to emotion recognition or biometric categorization. Deployers must tell people when these systems are used on them. In workplaces and educational settings, the AI Act goes further and prohibits emotion recognition outright, except for medical or safety reasons.
• Label deepfakes and certain AI-generated text. Deployers must disclose deepfakes, and AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and holds editorial responsibility.

For all four duties, the information must be clear and distinguishable, provided no later than a person's first interaction with or exposure to the system, and accessible. Thus, timing and format are part of the legal duty, which makes disclosure a true product design decision.

The Commission published its guidelines on July 20, 2026, following a Code of Practice on AI-generated content released in June. Neither is binding, but together they are the clearest guide to how regulators will apply Article 50.

The stakes are real. Member state authorities enforce Article 50, and under Article 99, breaches can bring fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher (for smaller companies, whichever is lower).

Dates that matter for assessment

Since the Digital Omnibus, the EU’s summer amendment package, entered into force on July 27, 2026, the timeline is:

• August 2, 2026: Article 50 transparency obligations apply, including informing people when they are interacting with AI.
• December 2, 2026: Machine-readable marking becomes required for generative systems that were already on the market before August 2. Systems launched since then have no grace period.
• December 2, 2027: Obligations for high-risk AI apply, including AI used in hiring, promotion, performance evaluation, and education. These include new duties to tell people when high-risk AI is used in decisions about them.

Why transparency matters most when AI measures people

A chatbot that helps you book a flight is one thing. An AI system that helps individuals and organizations understand skills, identify development opportunities, or evaluate proficiency carries different expectations for transparency. When the output shapes someone's career, disclosure stops being a checkbox and becomes the foundation of trust. 

That trust has practical consequences:

• Participation. Explaining how an assessment works makes it feel fairer and motivates people to perform, a consistent finding in research on how people react to assessments.
• Data quality. Disclosure can change how people respond. Telling participants that AI scores their work may lead some to write for the model rather than show their skill, so the most useful disclosure explains what is being measured, not how to game it.
• Defensibility. When an assessment result is questioned by a participant, customer or regulator, clear and timely disclosure provides a stronger foundation for transparency and accountability.  

The EU is not alone. Our team closely tracks AI and assessment regulation globally, and the pattern is consistent: across very different legal systems, and even where broader AI rules have been scaled back, the requirement to tell people when AI is involved keeps appearing.

• South Korea's AI Basic Act, in force since January 22, 2026, requires operators of high-impact AI, a category that includes hiring, to notify people in advance that AI is being used. It also applies to companies outside Korea whose AI affects Korean users. (Littler)
• Since January 1, 2026, Ontario has required employers to state in public job postings whether they use AI to screen, assess, or select applicants. That puts disclosure before a candidate ever applies. (Norton Rose Fulbright)
• Colorado replaced its comprehensive AI Act with a narrower, notice-based law. From January 1, 2027, people must be told when automated systems materially influence consequential decisions about them, including in employment and education, and they receive a disclosure and the option of human review after an adverse outcome. (Seyfarth Shaw).
• New York City's Local Law 144 requires employers using automated tools in hiring or promotion decisions to notify candidates in advance and to publish the results of an independent bias audit. Enforcement is tightening after a 2025 state audit found it had been weak. (Office of the NYS Comptroller)
• Professional standards point the same way: SIOP's 2023 recommendations hold AI-based assessments to the same validity and fairness evidence expected of any selection procedure.

The direction of travel is clear: people must be told before AI is used on them, or at the latest when they first encounter it. That makes timing a design decision.

How Workera puts transparency into the experience

Workera now discloses AI use at first touch and wherever people interact with it, across the participant, admin, assessment, and learning experiences. The update supports Article 50’s interaction-disclosure requirement and was shaped directly by enterprise customers asking for it as part of their Responsible AI reviews.

Here is what changed:

• An AI disclosure at first touch. Before they begin, users see and acknowledge a notice explaining how Workera uses AI.
• Built into onboarding. Leader and participant onboarding both explain the role AI plays, so the administrators and managers who deploy Workera understand it before their people encounter it.
• Disclosures in the moment. AI notices now appear in conversations with Elo, Workera's AI agent, and remain available from the user menu, so the information is there when people need it, not buried in a policy page.
• Coverage where the stakes are highest. Disclosures extend to hiring and pre-hire assessments and to proctored evaluations, where people have the most riding on the result. Hiring is also a use the AI Act will treat as high-risk from December 2027.
• Disclosures in learning, too. People building skills see the same AI notices as people being assessed, so the standard holds when the stakes are lower.

Disclosure is one part of Article 50. Next, Workera will continue to evaluate applicable Article 50 transparency requirements, including machine-readable marking of AI-generated content where required. 

Just as important is what the update does not do: it does not change how Workera's AI works. Workera assessments still measure verified proficiency, and the scoring behind decision-grade results is unchanged. 

That matters because a skills signal is defensible only when the evidence behind it holds up and the people it describes understand how it was produced. Being upfront about AI is part of what makes Workera defensible by design.

Six questions to ask your skills platform

If AI is part of how you hire, assess, or develop people, put these questions to every vendor in your stack:

1. Where in the product are users told that AI is involved, and can we see it?
2. Are candidates in hiring and proctored assessments covered, not just employees?
3. How does your platform help our administrators understand the AI before our people encounter it?
4. Do disclosures appear in the moment of interaction, or only in onboarding and legal terms?
5. How do you ensure AI disclosures remain clear, accessible and consistent as your platform evolves? 
6. What validity and fairness evidence stands behind your AI-based assessments?

The law sets a floor for telling people when AI is involved, and that floor rises in 2027. The organizations that earn trust in the AI era will be the ones that go beyond the minimum, making it obvious to every participant when AI is at work and why the results deserve to be relied on.

Want to see how Workera handles AI disclosure in practice, or how verified skills data supports your Responsible AI review? Talk to our team.

This post reflects developments as of October 2026 and is not legal advice. Organizations should consult their own counsel.

‍