Good news (sort of)... The EU AI Act's big compliance deadline for hiring AI, originally August 2, 2026, just got pushed to December 2, 2027. The EU finalized that delay in June 2026. One thing didn't move: the Act's first transparency rules still start August 2. For many hiring tools, especially anything a candidate interacts with directly, like an AI interviewer or chat-based assessment,  that means candidates need to be told AI is involved. If you've seen headlines treating August 2 as the day everything changes, they're out of date. The real story is simpler and more useful: you got extra time, and it's worth spending it well.

So what actually happens on August 2?

The EU AI Act has been rolling out in stages since 2024. Prohibited practices (things like social scoring) started in February 2025. Rules for general-purpose AI models started in August 2025. The part everyone's been watching, the high-risk rules that cover hiring, recruitment, and employment decisions, was supposed to start August 2, 2026. That date has now been pushed back.

Here's how that happened. Back in November 2025, the European Commission proposed a delay, worried that the technical standards and enforcement bodies needed to actually run this rule weren't ready in time. On May 7, 2026, the Council and Parliament agreed on that delay. Parliament signed off on June 16, and on June 29 the Council of the EU gave its final green light, fixing the new dates at December 2, 2027 for stand-alone high-risk systems and August 2, 2028 for high-risk AI embedded in products. So this isn't a rumor or a proposal anymore. Both co-legislators have signed off. The one step left is procedural: publication in the Official Journal, which the Council says will happen shortly, with the law entering into force three days after that, comfortably before August 2. Hiring, scoring, promotion, and termination AI systems now have until December 2, 2027 to comply.

One piece didn't move. Article 50, the transparency rule, stays on schedule for August 2, 2026. Its clearest application is to AI systems people interact with directly, such as  an AI interviewer, a chatbot screener, a conversational assessment. For those, candidates need a clear notice that they're dealing with AI. For backend tools that score or rank without the candidate ever talking to the AI, the picture is less settled: the fuller transparency duties for those systems arrive with the high-risk rules in 2027, and GDPR already requires informing people about automated decision-making in the meantime. Where exactly your tool lands is a question for counsel, but disclosing clearly is good practice either way, and costs you nothing.

Why the confusion happened

Honestly, it's an easy date to get wrong. EU regulation moves through a lot of stages, proposal, political agreement, formal adoption, publication, entry into force, and each one sounds like "it's happening" even when it isn't final yet. A proposal from last November got talked about as if it were already law. A political agreement in May got treated as the finish line, when it still needed sign-off from both Parliament and Council. As of this week, that sign-off is done. What's left is procedural: publication in the Official Journal, which the Council itself says is coming shortly. That's a real distinction (adopted versus in force), but not one that changes the outcome here.

What Article 50 actually asks for

If candidates interact with AI directly during your process, for example,  an AI-led interview, a chat-based screener,  they need to be told that clearly, before or during the interaction. That's the core of what's due August 2. Whether purely backend scoring tools also trigger the notice this summer is genuinely debatable, and worth a conversation with employment counsel rather than a guess. What's not debatable: you don't need to prove the tool is fair, document an override process, or register anything with the EU. Those come later, under the high-risk rules, now due in 2027.

That distinction is worth sitting with for a second. Telling someone AI is involved and proving that AI is fair are very different jobs. The second one is the harder, more important one, and it's the one that actually protects candidates. A wave of "AI Act compliant" claims based only on a disclosure notice would be missing the point. Good assessment science, the kind with real validity evidence, has been solving the harder problem for years already. Decades of selection research show that structured, validated methods consistently outpredict instinct and unstructured judgment when it comes to who will actually perform. That's the bar worth aiming for, regardless of what any regulation requires by when.

Three things worth doing now anyway

None of these depend on the calendar. They're just good practice, whether the deadline is this summer or 2027.

• Get the disclosure right, and get it right now. Put a clear, plain-language notice in the candidate flow before August 2, not buried in a terms-of-service page nobody reads. If candidates interact with AI directly, it's required. If your AI works behind the scenes, the legal picture is fuzzier, but disclosing anyway is simple, builds trust, and future-proofs you for 2027, when transparency duties expand. 
• Keep building your evidence base, even without a deadline forcing it. Adverse impact data, validity evidence by role and group, and a clear human oversight process are worth having whether or not a regulator asks for them yet. Research on AI hiring tools published in the Journal of Management Studies has found that standardizing a process can fix one fairness problem while quietly introducing another, which is exactly the kind of thing you only catch if you're actually tracking outcomes.
• Get ready for buyer questions before buyers ask them. The delay means every RFP from now to 2027 will include AI Act questions, and the vendors with crisp answers will win deals the vague ones lose. Compliance readiness is about to become a sales asset.

One caveat worth stating: this is guidance, not legal advice. How the Act applies depends on your tools and jurisdiction, so have qualified employment counsel confirm your specific position.

Where Deeper Signals stands

Deeper Signals builds hiring, development, and team assessments, grounded in industrial-organizational psychology. For HR teams, that means scores backed by published validity evidence rather than a black box. For candidates, it means a clear, explainable process, with results framed as useful development input rather than a pass/fail verdict. And for anyone watching this regulation closely, it means a platform built to hold up to real scrutiny well before the Act made that a legal requirement. Assessment data stays inside Deeper Signals and is never used to train outside models.

There's a tendency to treat the regulatory deadline as the finish line for "responsible AI in hiring." We'd push back on that a little. Good I/O psychology has required things like validity evidence, adverse impact tracking, and human oversight for decades, long before any regulator asked for them. A fairness case that only exists because a law demanded it was never going to be a strong one. This delay doesn't lower the bar. It just gives everyone more time to clear the one that was already there.

FAQ

1. Is the August 2, 2026 deadline still happening? 

Partly. Article 50's transparency rules still start August 2. They apply most clearly to AI that candidates interact with directly. The bigger high-risk rules covering hiring AI overall have been pushed to December 2, 2027. If you're unsure which bucket your tools fall into, ask for legal advice.

2. Why did it get delayed? 

The EU didn't have the technical standards or enforcement bodies ready in time, so lawmakers agreed to give everyone more runway.‍

3. What should I ask my assessment vendor right now? 

Two things: is the AI disclosure notice live in their product today, and what's their plan for the 2027 requirements.

4. Does this apply if my company isn't based in the EU? 

Possibly, yes. Like GDPR, the Act applies based on where the effects land, not where your office is. A US company screening candidates for an EU role can still be in scope.