Hiring AI is high-risk.Talara is built for it.

The deadline moved. The obligation didn’t.

On 27 July 2026, the EU’s Digital Omnibus (Regulation (EU) 2026/1744) moved the Annex III high-risk deadline for hiring and HR systems from 2 August 2026 to 2 December 2027. If your hiring AI screens, ranks, scores, or selects candidates, here’s what actually changed, and what didn’t.

• The heavy obligations, deferred. Documented risk management, conformity assessment, and registration in the EU database now have until 2 December 2027, sixteen months later than originally set.
• Transparency and AI literacy, unchanged. Article 50’s duty to tell candidates when AI is involved in a decision about them, and Article 4’s duty to make sure the people running the system understand it, were not delayed. Both are already how Talara works.
• The extension is runway, not a reason to wait. Retrofitting human-oversight gates, bias monitoring, and an audit trail onto a hiring process that wasn’t built for them is a product change, not a policy update. Sixteen extra months is time to do it properly, not time to leave it until 2027.

What the EU AI Act requires for hiring

The EU AI Act (Regulation (EU) 2024/1689) lists AI systems used for the recruitment or selection of candidates among its high-risk categories. For a deployer, that brings a set of obligations:

• Human oversight: a person must be able to understand, oversee, and override the system.
• Transparency: people affected must be informed, and outputs must be interpretable.
• Risk and bias management: foreseeable risks, including discrimination, must be identified and mitigated.
• Record-keeping: the system must log events so decisions can be traced after the fact.
• Data governance: data used must be relevant and managed lawfully.

How Talara maps to each obligation

Talara is designed so these obligations are satisfied by the product itself, not by process documents alone:

• Human oversight is enforced as a database constraint: no candidate advances, no offer sends, and no role closes without a recorded human approval.
• Transparency is built into every AI output: each fit score and recommendation shows the reasoning behind it.
• Bias monitoring runs across the pipeline so disparate outcomes surface early.
• Record-keeping is the immutable audit trail: every action carries the actor, the subject, and the before-and-after change.

Is your hiring AI in scope?

If a system screens, ranks, scores, or selects candidates, Annex III captures it: there’s no small-company carve-out, and matching features inside a traditional ATS count, not just standalone AI products. The obligations that follow (human oversight, transparency, risk and bias management, record-keeping, data governance) rest with you as the deployer, not only with your vendor.

Does this apply outside the EU?

The Act’s reach doesn’t stop at EU borders: it applies when hiring AI is used for candidates in the EU, wherever the company running it sits. If you hire into an EU office, plan as if you’re in scope, and confirm the specifics with counsel.

What you can show an auditor

Because every consequential decision is gated and logged, Talara lets you produce the evidence an auditor or works council asks for: who made each decision, what the AI proposed, what reasoning it showed, and what the human did with it. The record is exportable and keyed to resource identifiers rather than personal data, so it survives GDPR erasure requests.

This is the same model described under human-in-the-loop hiring.

The full picture of how the gate works day to day, and what it hands back to your team, is on the human-in-the-loop page.