The EU AI Office reportedly released practical guidelines this week clarifying Article 6 of the AI Act, the provision that determines which systems must meet high-risk compliance requirements. According to coverage of the guidance, the document gives explicit treatment to autonomous agents operating in HR and credit scoring contexts, categorizing them as high-risk on the basis of their potential for “significant influence” on fundamental rights.
That categorization isn’t new in principle. EU AI Act Annex III has listed employment and HR management systems, and systems used to evaluate creditworthiness, as high-risk categories since the regulation passed in 2024. What the reported guidance adds is practical specificity: according to reports, it directly addresses how autonomous agents in those contexts should be treated, a question the Act’s text left open to interpretation for many deployers.
The guidance reportedly also addresses logging requirements, citing obligations across four articles of the Act. Logging provisions appear throughout the EU AI Act’s text, Articles 12, 17, 26, and 73 each contain record-keeping and event-logging obligations for high-risk systems, so the “four articles” characterization is consistent with the Act’s structure. Specific event-logging requirements as the guidance frames them haven’t been confirmed from an authoritative source yet, and the full guidance document warrants direct review.
The urgency here is the calendar. August 2, 2026 is the date by which Annex III providers and deployers must meet technical documentation, conformity assessment, logging, and human oversight requirements. That date comes from the Act’s official enforcement timeline, confirmed in the regulation’s text and corroborated by multiple compliance planning publications. Four months is not much runway for organizations that haven’t started.
A proposal to push that deadline to 2027 is circulating under the Digital Omnibus package, but it hasn’t been enacted. Compliance teams planning around a 2027 date are betting on a proposal that has no guarantee of passage. The EU AI Office issuing practical classification guidance while the delay proposal is still unresolved sends a clear institutional signal: the Office is treating August 2 as real.
What to watch: Whether the guidance document surfaces publicly with an official EU domain URL, which would upgrade the current “reportedly” framing on specific provisions to confirmed requirements. The Digital Omnibus delay proposal timeline is the other moving piece. If it advances materially before summer, enforcement planning calculus changes. If it stalls, August 2 becomes a hard stop for any organization that hasn’t mapped its systems against Annex III.
The TJS read: This guidance matters most to two audiences. First, HR technology vendors and enterprise HR teams that have deployed or are deploying AI in recruiting, performance evaluation, or workforce management, especially any system that operates with meaningful autonomy. Second, fintech and credit platforms with EU user exposure. Both groups should pull their Annex III mapping work forward now rather than wait for the guidance document to be widely distributed through compliance channels. Waiting on the Digital Omnibus delay is a planning risk, not a planning strategy.