[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fAPh-kCJeUQzJl2LknCttFg-945Bv8QWvjEKP7f_nNd4":3,"$fm6Ky0r9HHO8PTXKBgXJeU4fCLm4z6RKRMXFRXG3f560":29},{"success":4,"data":5},true,{"id":6,"slug":7,"title":8,"excerpt":9,"content":10,"category":11,"tags":12,"author":19,"cover_image_url":20,"reading_time_minutes":21,"is_published":4,"published_at":22,"created_at":23,"updated_at":24,"author_avatar":25,"is_featured":26,"meta_title":27,"meta_description":28,"meta_keywords":25},"6590a54f-be79-4856-baec-5328a68a5442","eu-ai-act-august-2026-what-applies-hiring","2 August 2026 Came and Went: What the EU AI Act Now Actually Requires of AI Hiring","Six days before the EU AI Act's high-risk regime was due to hit recruitment AI, the deadline moved to 2 December 2027. But Article 50 transparency did take effect on 2 August 2026, and the workplace emotion-inference ban has been live since 2025. What moved, what didn't, and what hiring teams should do with a sixteen-month reprieve.","\u003Cp>For two years, 2 August 2026 was the date on every AI-hiring compliance roadmap. It was the day the EU AI Act's high-risk regime — the one that names recruitment and selection systems explicitly — was scheduled to start applying. Vendor questionnaires were built around it. Budget cycles were argued from it. Then, six days before the deadline, the deadline moved.\u003C\u002Fp>\n\n\u003Cp>On 24 July 2026 the EU published the \"Digital Omnibus on AI\" — Regulation (EU) 2026\u002F1744, the first formal amendment package to the AI Act — in the Official Journal, and it entered into force on 27 July, per the analysis published that day by employment law firm \u003Ca href=\"https:\u002F\u002Fwww.lewissilkin.com\u002Finsights\u002F2026\u002F07\u002F27\u002Fthe-digital-omnibus-on-ai-enters-into-force-today-102nedo\" rel=\"noopener\" target=\"_blank\">Lewis Silkin\u003C\u002Fa>. The headline change: the obligations for stand-alone high-risk AI systems, which include AI used for recruitment, no longer apply from 2 August 2026. They apply from \u003Cstrong>2 December 2027\u003C\u002Fstrong>.\u003C\u002Fp>\n\n\u003Cp>So last Saturday passed quietly. But \"quietly\" is not \"nothing happened\" — parts of the Act \u003Cem>did\u003C\u002Fem> start applying on 2 August, and one of them is the part your candidates will actually notice. This piece separates what moved, what did not, and what a hiring team should do with a sixteen-month reprieve. As always, we keep three things distinct: \u003Cstrong>what the law says\u003C\u002Fstrong>, \u003Cstrong>what remains genuinely unsettled\u003C\u002Fstrong>, and \u003Cstrong>our own reading\u003C\u002Fstrong> — and we label which is which.\u003C\u002Fp>\n\n\u003Ch2>What moved: the high-risk regime, in two tranches\u003C\u002Fh2>\n\n\u003Cp>Under the original Act, AI systems used for \"recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates\" sit in Annex III — the high-risk list — alongside systems used for promotion, termination, task allocation and monitoring decisions at work. The full obligations package for those systems (risk management, data governance, technical documentation, human oversight, conformity assessment, registration) was due on 2 August 2026.\u003C\u002Fp>\n\n\u003Cp>The Omnibus defers that in two tranches, as summarized consistently by \u003Ca href=\"https:\u002F\u002Fwww.gibsondunn.com\u002Feu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes\u002F\" rel=\"noopener\" target=\"_blank\">Gibson Dunn\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fdatamatters.sidley.com\u002F2026\u002F06\u002F22\u002Feu-lawmakers-reach-provisional-agreement-to-delay-key-eu-ai-act-obligations\u002F\" rel=\"noopener\" target=\"_blank\">Sidley\u003C\u002Fa>:\u003C\u002Fp>\n\n\u003Cul>\n\u003Cli>\u003Cstrong>Stand-alone Annex III high-risk systems\u003C\u002Fstrong> — the category almost all hiring AI falls into — now become subject to the high-risk obligations on \u003Cstrong>2 December 2027\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>AI embedded in products already regulated under EU product-safety law\u003C\u002Fstrong> (Annex I — think AI-enabled medical devices, not HR software) follows on \u003Cstrong>2 August 2028\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\u003Cp>One detail worth pausing on. The European Commission's original proposal last November tied the delay to a \u003Cem>conditional\u003C\u002Fem> trigger — obligations would bite once harmonized standards were available. The final text dropped that mechanism in favour of \u003Cstrong>fixed dates\u003C\u002Fstrong>. Our reading: that is the more employer-friendly outcome, not because it is later, but because it is \u003Cem>plannable\u003C\u002Fem>. A conditional trigger would have meant compliance teams watching standards-body publication schedules to learn their own deadline. A fixed date, sixteen months out, is something a procurement cycle can actually be built around.\u003C\u002Fp>\n\n\u003Ch2>What did not move — and is live right now\u003C\u002Fh2>\n\n\u003Cp>Three things matter here, and none of them was deferred.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>1. Article 50 transparency applies as of 2 August 2026.\u003C\u002Fstrong> The duty to ensure that people are told when they are interacting with an AI system — unless it is obvious from context — took effect on schedule. Both Gibson Dunn and Lewis Silkin note that Article 50 was left essentially untouched by the Omnibus. For hiring, this is the provision with a face: an AI chatbot answering candidate questions, an AI voice agent conducting a screening call, an avatar-fronted video interview — these are AI systems interacting directly with natural persons. The one grace period the Omnibus added here is narrow: machine-readable marking of synthetic content gets until \u003Cstrong>2 December 2026\u003C\u002Fstrong>, and only for systems already on the market before 2 August 2026. The disclosure duty itself is live.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>2. The prohibitions have been live since February 2025 — including emotion inference at work.\u003C\u002Fstrong> Article 5(1)(f) prohibits AI systems that \"infer emotions of a natural person in the areas of workplace and education institutions\", with a carve-out for medical or safety purposes, and it has applied since \u003Cstrong>2 February 2025\u003C\u002Fstrong> (\u003Ca href=\"https:\u002F\u002Fartificialintelligenceact.eu\u002Farticle\u002F5\u002F\" rel=\"noopener\" target=\"_blank\">full article text\u003C\u002Fa>). The Omnibus did not touch it. This matters for hiring more than the abstract wording suggests: video-interview analytics that claim to read enthusiasm, stress or engagement from a candidate's face or voice are exactly the kind of feature this provision was written about. If a tool in your stack advertises emotion analytics for interviews in the EU, the deferral of the high-risk regime does not help it — that question is not deferred, it is prohibited territory with a narrow exception.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>3. The scaffolding kept arriving.\u003C\u002Fstrong> Governance structures, the general-purpose AI obligations that began in August 2025, and Member State penalty frameworks were not part of the deferral. The machinery that will enforce the December 2027 obligations is being assembled now, on the original schedule.\u003C\u002Fp>\n\n\u003Ch2>The quieter changes hiring teams should know about\u003C\u002Fh2>\n\n\u003Cp>The Omnibus was not only a calendar edit. Three substantive changes are worth a hiring lens:\u003C\u002Fp>\n\n\u003Cul>\n\u003Cli>\u003Cstrong>AI literacy (Article 4) was softened.\u003C\u002Fstrong> The obligation shifts from organisations having to \"guarantee a sufficient level\" of AI literacy toward documented training and awareness efforts, with the Commission and Member States taking on a support role. Deployers of high-risk systems still face specific competence requirements under Article 26(2). Our reading: the paperwork burden shrank; the underlying expectation — that the recruiter overriding an AI ranking understands what the tool does and where it fails — did not.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>A narrowed \"safety component\" definition.\u003C\u002Fstrong> AI performing \"non-safety related aspects of user assistance, performance optimization, service efficiency, automation or convenience\" no longer automatically triggers high-risk status when embedded in regulated products, unless failure would endanger health or safety. This mostly matters outside HR, but it tightens the question every vendor will now be asked: \u003Cem>which\u003C\u002Fem> limb of the Act does your product actually fall under?\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Registration was simplified, not removed\u003C\u002Fstrong>, for systems self-assessed as non-high-risk under Article 6(3). Expect some hiring-tech vendors to argue their tools fall in that bucket. The registration trail — and the self-assessment behind it — is precisely where those arguments will be tested.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\u003Ch2>Our reading: this is a deadline change, not a direction change\u003C\u002Fh2>\n\n\u003Cp>Here is the part we would put in front of any talent-acquisition lead deciding what to do with the news.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>The substance is untouched.\u003C\u002Fstrong> Recruitment AI is still named, still high-risk, still headed for the full obligations package. Nothing in the Omnibus re-opened \u003Cem>whether\u003C\u002Fem> hiring AI is high-risk — only \u003Cem>when\u003C\u002Fem> the obligations attach. Sixteen months is roughly one enterprise procurement cycle. A team that selects an AI hiring tool in early 2027 on a three-year contract is selecting the tool it will have to make compliant in December of that same year.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>The candidate-facing duty is already live.\u003C\u002Fstrong> Disclosure is the obligation with the shortest distance between non-compliance and reputational damage, because the person it protects is the person you are trying to hire. A candidate who discovers mid-process that the \"interviewer\" was an AI nobody told them about does not file a regulatory complaint first — they post about it.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>The rest of the world did not defer.\u003C\u002Fstrong> The EU moved a date; other regimes kept moving. In our own home market, Hong Kong's Privacy Commissioner has been running AI compliance checks against employers under the PDPO, with the Model AI Framework explicitly naming assessment of job applicants as a higher-risk use — we covered that in detail \u003Ca href=\"https:\u002F\u002Fwww.nicehire.ai\u002Fblog\u002Fhk-pcpd-model-ai-framework-hiring\">here\u003C\u002Fa>. A hiring stack that treats the EU deferral as a general permission slip will be wrong-footed somewhere else.\u003C\u002Fp>\n\n\u003Cp>So the practical list for the next quarter looks like this — and it is deliberately boring:\u003C\u002Fp>\n\n\u003Col>\n\u003Cli>\u003Cstrong>Inventory\u003C\u002Fstrong> every AI touchpoint in your hiring flow, including the ones inside vendor products (ranking, parsing, chat, voice, video analysis). You cannot assess what you have not listed.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Check disclosure today\u003C\u002Fstrong>, not in 2027. Walk your own candidate journey and note every point where a person interacts with a machine without being told. Article 50 is live; this is the audit you can do in an afternoon.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Ask vendors the emotion question directly.\u003C\u002Fstrong> Does any feature infer emotional state from candidates? If yes, in the EU that is an Article 5 conversation, not an Annex III one.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Keep building the documentation spine\u003C\u002Fstrong> — risk assessment, human-oversight design, records of what the system was asked to do. Every serious analysis of the Omnibus, including Lewis Silkin's, lands on the same advice: the deferral rewards teams that use the time, not teams that stop.\u003C\u002Fli>\n\u003C\u002Fol>\n\n\u003Ch2>Where NiceHire stands\u003C\u002Fh2>\n\n\u003Cp>A short, factual note on our own product, since we run AI screening interviews and it would be odd not to say where we sit. Every NiceHire AI screening greeting identifies the interviewer as an AI — in English, Japanese, and Traditional and Simplified Chinese, on first attempt and on reconnect after a dropped call. A customer or template greeting can change the wording and tone, but cannot remove the disclosure: every greeting-producing path passes through a control that adds a one-sentence disclosure in the greeting's language if it is missing. We built it that way not because a specific regulation currently compels us to, but because a candidate should be able to tell they are talking to a machine. Completed AI screening interviews also record an interview transcript against the application — with two honest limits: the transcript is what the interview provider returns, so a session that ends without one is stored without one, and re-running a screening currently replaces the previous attempt's transcript rather than versioning it.\u003C\u002Fp>\n\n\u003Cp>That is the same standard we would suggest applying to any vendor, including us: ask what the product does today, in the tense the code supports — not what the roadmap promises for December 2027.\u003C\u002Fp>\n\n\u003Chr \u002F>\n\n\u003Cp>\u003Cem>Sources, all accessed 4 August 2026: \u003Ca href=\"https:\u002F\u002Fwww.lewissilkin.com\u002Finsights\u002F2026\u002F07\u002F27\u002Fthe-digital-omnibus-on-ai-enters-into-force-today-102nedo\" rel=\"noopener\" target=\"_blank\">Lewis Silkin — \"The Digital Omnibus on AI enters into force today\"\u003C\u002Fa>; \u003Ca href=\"https:\u002F\u002Fwww.gibsondunn.com\u002Feu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes\u002F\" rel=\"noopener\" target=\"_blank\">Gibson Dunn — \"EU AI Act Omnibus Agreement\"\u003C\u002Fa>; \u003Ca href=\"https:\u002F\u002Fdatamatters.sidley.com\u002F2026\u002F06\u002F22\u002Feu-lawmakers-reach-provisional-agreement-to-delay-key-eu-ai-act-obligations\u002F\" rel=\"noopener\" target=\"_blank\">Sidley Data Matters — \"EU Lawmakers Reach Provisional Agreement to Delay Key EU AI Act Obligations\"\u003C\u002Fa>; \u003Ca href=\"https:\u002F\u002Fartificialintelligenceact.eu\u002Farticle\u002F5\u002F\" rel=\"noopener\" target=\"_blank\">EU AI Act, Article 5 (full text)\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>","Compliance",[13,14,15,16,17,18],"EU AI Act","compliance","AI hiring","Digital Omnibus","regulations","transparency","NiceHire Team","https:\u002F\u002Fsomdgtvycghtiuhiyaly.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fnicehire\u002Fblog-covers\u002Feu-ai-act-august-2026-what-applies-hiring.webp",8,"2026-08-04T06:06:52.887+00:00","2026-08-04T06:06:53.110375+00:00","2026-08-04T06:22:55.936972+00:00",null,false,"EU AI Act After 2 August 2026: What Applies to AI Hiring Now","The Digital Omnibus deferred the EU AI Act's high-risk obligations for recruitment AI to December 2027 — but AI disclosure duties took effect 2 August 2026.",{"success":4,"data":30},{"posts":31,"count":58,"hasMore":26},[32,34,47],{"id":6,"slug":7,"title":8,"excerpt":9,"category":11,"tags":33,"author":19,"cover_image_url":20,"reading_time_minutes":21,"published_at":22},[13,14,15,16,17,18],{"id":35,"slug":36,"title":37,"excerpt":38,"category":11,"tags":39,"author":19,"cover_image_url":25,"reading_time_minutes":45,"published_at":46},"380f0e8c-c2c8-42c2-af32-bcee561e897a","hk-pcpd-model-ai-framework-hiring","Voluntary on Paper, Benchmark in Practice: What the PCPD's Model AI Framework Means for Hiring in Hong Kong","Hong Kong has no AI statute, but the PDPO binds every employer touching candidate data — and the PCPD's Model AI Framework names the assessment of job applicants as a higher-risk AI use. What the Framework says, how it maps onto a recruitment workflow, and what three annual rounds of published compliance checks signal.",[40,41,42,15,14,43,44],"PCPD","PDPO","Hong Kong","Model AI Framework","data protection",10,"2026-07-24T01:37:25.988+00:00",{"id":48,"slug":49,"title":50,"excerpt":51,"category":11,"tags":52,"author":19,"cover_image_url":25,"reading_time_minutes":56,"published_at":57},"8be0cb6b-6afa-43fa-9798-52127f706332","eu-ai-act-article-50-explainer","The EU AI Act Delay Did Not Cover AI Interviews. Article 50 Applies on 2 August 2026.","The Digital Omnibus pushed the AI Act's high-risk hiring obligations to December 2027 — but the Article 50 duty to tell a candidate they are talking to a machine still applies from 2 August 2026. What moved, what didn't, who carries the obligation, and where NiceHire's own code stood when we checked.",[13,53,54,14,18,16,55],"Article 50","AI interviews","hiring",11,"2026-07-24T01:15:53.793+00:00",3]