Deferred is not exempt — Adecco puts candidate-selection agents in 40 countries, and the clock that moved is not the one that binds

On 15 September 2026 the Adecco Group said it is rolling Salesforce's Agentforce Coworker, powered by Anthropic's Claude, to 27,000 employees across 40 of the 60 countries it operates in, with recruitment agents that select candidates and launch pre-screening. The obvious reading is that the Digital Omnibus bought everyone time: Regulation (EU) 2026/1744, in force since 27 July 2026, moved Annex III high-risk obligations to 2 December 2027. The regulation's own title corrects it — it amends Regulations (EU) 2024/1689, 2018/1139 and 2023/1230, and not Regulation (EU) 2016/679. Candidate selection is Annex III point 4(a) verbatim, the Article 6(3)(d) preparatory-task derogation is closed by the same paragraph's rule that profiling is always high-risk, and the GDPR duty that binds today has been in force since 2018: in Case C-634/21 the Court held that an automated probability value is itself the Article 22 decision where the recipient draws strongly on it. The practitioner move is an Article 22 inventory taken off one agent trace rather than an AI Act gap analysis dated 2027.

Post

Adecco is putting candidate-selection agents into forty countries. The deadline everyone will quote at you is the wrong one.

On 15 September the Adecco Group said it is rolling Salesforce's Agentforce Coworker — powered by Anthropic's Claude — to 27,000 employees across 40 of the 60 countries it operates in. In recruitment the agents select candidates and launch pre-screening, recruiter and onboarding agents. Ten countries already run agentic AI in recruitment, and those ten are half of group revenue. 2.5 million agent-candidate interactions since April 2025.

What this changes for anyone shipping agents in the EU:

  1. There are two clocks and only one of them moved. Regulation (EU) 2026/1744, the Digital Omnibus on AI, in force 27 July 2026, pushed Annex III high-risk to 2 December 2027. Read its title: it amends Regulations 2024/1689, 2018/1139 and 2023/1230. Not 2016/679.

  2. The GDPR half is still a proposal. The Article 22 rewrite that would make a solely-automated decision ordinarily lawful sits in the data omnibus, still in negotiation. You cannot build a plan on it.

  3. Candidate selection is Annex III point 4(a) in the Act's own words: systems used "to analyse and filter job applications, and to evaluate candidates". That is not a reading. That is the text.

  4. The narrow-task escape does not open. Article 6(3)(d) exempts a system performing "a preparatory task to an assessment" — which is exactly what pre-screening sounds like. Same paragraph, last sentence: a system that "performs profiling of natural persons" is always high-risk. Pre-screening profiles.

  5. The Court got there in 2023. In C-634/21 the CJEU held that an automated probability value is itself the Article 22 decision, where the party who receives it draws strongly on it. Your agent never has to reject anyone. It only has to be believed.

My Monday: open one agent run and mark every step where it orders, scores or drops a candidate. That list is your Article 22 inventory and your Article 35 DPIA trigger, and it is dated today rather than December 2027.

Deferred is not exempt. Which clock is your roadmap on?

#DSGVO #AIAct #AgenticAI #HRTech #Compliance

AI disclosure: the narration voice, the cover art and the brand ident animation in this video are AI-generated. The script, the claims and the source checks are mine.

First comment

Sources, in the order the argument uses them.

The Adecco Group's own announcement, 15 September 2026 — 40 of 60 countries, 27,000 employees, the recruitment agent list, and the 2.5 million agent-candidate interactions: https://www.adeccogroup.com/our-group/media/press-releases/adecco-group-salesforce-agentforce-40-countries

Salesforce's AIforce announcement, the same day — what Agentforce Coworker is and that it arrives in the Lightning interface at the push of a button: https://www.salesforce.com/news/stories/aiforce-announcement/

Regulation (EU) 2026/1744, the Digital Omnibus on AI — the point is in the title: it amends 2024/1689, 2018/1139 and 2023/1230, and nothing else: https://eur-lex.europa.eu/eli/reg/2026/1744/oj

White & Case on the AI Omnibus — the moved dates in one place, and the confirmation that the GDPR amendments remain under negotiation: https://www.whitecase.com/insight-alert/eu-ai-omnibus-enters-force-amending-ai-act

AI Act Annex III, point 4(a) — recruitment and selection: https://artificialintelligenceact.eu/annex/3/

AI Act Article 6(3) — the four derogations, and the profiling sentence that closes them: https://artificialintelligenceact.eu/article/6/

CJEU, Case C-634/21, SCHUFA Holding (Scoring), judgment of 7 December 2023 — the automated probability value as the Article 22 decision: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A62021CJ0634

Article 22 GDPR, unchanged since 2018: https://gdpr-info.eu/art-22-gdpr/

Cut for length: Article 26(7) of the AI Act makes a deployer inform workers' representatives before putting a high-risk system into service at work. That duty really is dated December 2027 — and it is the one that will surprise people, because it is not a filing, it is a conversation.

AI disclosure: the narration voice, the cover art and the brand ident animation in this video are AI-generated. The script, the claims and the source checks are mine.

Claims

Sources

Corrections

None.

AI disclosure

The video for this edition contains AI-generated media: the narration is a synthetic voice, the cover art and any background imagery are generated, and the brand ident animation is generated. The script, the claims and the source verification are human-authored and human-checked. Assets carry IPTC DigitalSourceType and C2PA Content Credentials.