Interim order on Meta's assistants; agents that escaped testing; Council settles its Omnibus line

AI Office guidelines page — One Commission page now tracks every AI Office guidance document, published and pending. Out already: the high-risk guidelines for providers and deployers, the transparency guidelines, the serious-incident reporting template, and practical guidance on the high-risk requirements. Queued: a fundamental rights impact assessment template, value-chain responsibilities, substantial modification, simplified quality-management guidance for SMEs — plus joint AI Act–GDPR guidelines with the EDPB. Your single checklist of what guidance exists before you brief anyone on obligations. Stakeholders have pressed the Commission to clarify the research exemptions in Articles 2(6) and 2(8). European Commission

External evaluator requirements — The AI Office has published its workshop report on outside evaluators for GPAI models with systemic risk — the Article 55(1)(b) route — proposing a scoring system for evaluators' qualifications and their independence from provider influence. Providers of systemic-risk models now have a written reference for who can credibly test them, though the Office says its views may still evolve as evaluation practice develops. Nineteen organisations joined the 15 July workshop, from 20 selected after a public call. European Commission

Meta interim order — MLex reported on 21 October that the European Commission has published an interim order concerning Meta's AI assistants. Meta was among the developers the Commission questioned over the summer on the Act's safety and copyright rules. If you run Meta assistants in the EU, the order's operative terms matter to you; the details sit behind MLex's paywall, so watch for the Commission's own publication. Reported by MLex on 21 October 2026. MLex

AI Act Scientific Panel — The Commission has filled its Article 68 Scientific Panel with 60 independent experts on renewable 24-month terms (Implementing Regulation (EU) 2025/454). Chaired by Seán Ó hÉigeartaigh with Nicolas Moës as vice-chair, it advises the AI Office and national authorities and alerts the Office to systemic risks. It is the technical body behind the enforcement escalation above — it briefed Virkkunen on 9 October and supports market surveillance. Membership is capped at three nationals per country; Yoshua Bengio is among the experts. European Commission

Hamburg data protection authority — Hamburg's data protection authority will take over oversight of the city's public-sector AI from January, MLex reported on 21 October. Public-sector deployers in the city will face one regulator for both AI Act and GDPR questions from the new year. Worth a note in any German public-sector deployment map, as market surveillance takes shape. The article body is paywalled; only the headline facts are public. MLex

Digital Omnibus — The Irish Presidency has landed member states' negotiating position on the proposed Digital Omnibus (IAPP, 22 October). The Council keeps the GDPR's definition of personal data unchanged, sets out where cookie consent is unnecessary — service delivery, audience measurement, device security — and keeps incident reporting at national portals. For AI Act readers, the mandate permits sensitive data use in biometric verification and AI model training and clarifies lawful grounds for training data — watch this in trilogue. Trilogues await the Parliament's own position. IAPP

“simplify rules on data protection, privacy and cybersecurity for businesses, public administrations and citizens, while maintaining established policy goals.”

Hungarian non-paper — Hungary has circulated a non-paper proposing an EU Frontier AI Safety and Security Team — vetted evaluators, independent of the labs, working inside developers' training and testing operations from day one and reporting to the AI Office and national contact points. The paper also urges an extra capability tier in the Act for pre-market development and major internal deployment, plus liability rules for autonomous agents. A proposal, not law — but it shows where the frontier-AI debate in the Council is moving, and it would change how any EU lab builds frontier models. It also floats national contact points for loss-of-control incidents, a pooled EU evaluation capability, and an international track beginning with middle powers before widening to the US and China. The EU AI Act Newsletter (Future of Life Institute)

Solier — Solier, the Parliament's lead lawmaker on the file, has told MLex that the EU's sovereign cloud and AI legislation could be simplified. No text exists; this is an on-the-record statement of intent, with the detail behind MLex's paywall. Reported 26 October 2026. MLex