EU AI Act Delays
The EU’s AI Act has hit implementation delays, creating a temporary 'vulnerability window' for high‑risk systems — regulators are falling behind rapid hardware and AI deployments. At the same time, analysts warn agentic AI in finance raises moral‑hazard and auditability challenges that firms need to manage now. (techpolicy.press) (azorobotics.com)
IMCO and LIBE committees voted 101–9 (with 8 abstentions) on 19 March 2026 to adopt a negotiating position that fixes new high‑risk AI compliance dates to 2 December 2027 and a secondary August 2028 milestone. (europarl.europa.eu) (aiactblog.nl) The European Commission’s “Digital Omnibus” proposal, first published 19 November 2025, formally moved the proposal to delay the AI Act’s most onerous obligations for high‑risk systems from the original 2 August 2026 timetable to late‑2027. (euractiv.com) (euractiv.com) The Commission also missed a formal guidance deadline tied to Article 6 — guidance due 2 February 2026 on how operators should comply with high‑risk duties — leaving conformity assessment and enforcement playbooks incomplete during the interim. (iapp.org) (iapp.org) Civil‑society groups and watchdogs warn the postponement plus the “omnibus” text’s broader carve‑outs creates a de‑facto grandfathering window that could keep many already‑deployed high‑risk systems outside new conformity checks until December 2027. (techpolicy.press) (techpolicy.press) Industry and regulators are also struggling to match the pace of compute and hardware rollout: the European Court of Auditors’ 2024 review found the EU has underinvested in AI infrastructure and capacity, while expert proposals now call for compute‑level measures such as a global chip registry to track accelerator flows. (eca.europa.eu) (eca.europa.eu) Financial‑sector advisory notes now flag agentic AI as an immediate operational risk: Deloitte recommends treating AI agents as active operators that require explicit control, audit trails and human‑in‑the‑loop governance, while analyst reports document real incidents used to illustrate moral‑hazard and auditability failures. (deloitte.com) (deloitte.com) Credit and risk houses are translating those warnings into credit and governance guidance: Moody’s and several law‑firm whitepapers list agentic workflows (portfolio rebalancing, routing orders across venues) as use‑cases needing bespoke testing, control logs and liability allocation before firms scale them in live trading stacks. (moodys.com) (moodys.com)
