The AI Act’s hardest rules for finance have slipped to December 2027. The supervisors did not slip with them, and the one duty that kept its date is a duty to tell customers when AI is in the room.
This article was originally published in FinanceX Magazine Issue #28 (October 2026) as part of their Technology Frontiers edition. We’re grateful for the opportunity to share this perspective with the European fintech community. As part of our partnership with FinanceX Magazine, we’re pleased to share our article on what the AI Act delay changes for European finance: why explanation, not compliance, is the test that arrives first, and why that is a content question as much as a governance one.
Human-led, AI-assisted: this article was outlined, written and edited by Jeremy Rodgers, with AI tools (Claude Fable 5.1 by Anthropic, Gemini 3.6 Pro by Google) used for research, drafting and fact-checking.
What did European finance win on 27 July? On paper, sixteen months. The Digital Omnibus on AI entered into force that day and pushed the AI Act’s high-risk obligations back from 2 August 2026 to 2 December 2027 for stand-alone systems, the category that includes credit scoring and life and health insurance pricing (European Commission, 2026). The AI literacy duty, in force since February 2025, was eased as well.
Tempting to read that as time off. Read it instead as a change of date, not of question. Behind every clause of the AI Act sits one test: can you say, in plain words, what your AI does and who is watching it? Frankfurt was setting that test long before Brussels moved the deadline.
In February, the ECB’s Pedro Machado said that 85% of large banks under European supervision now use AI in some form, and that usage had shifted from experimentation to materiality (ECB, 2026). His measure of whether a bank controls a model is whether it can explain the model’s behaviour in terms that mean something to decision-makers. A bank that cannot, in his words, “cannot truly control that model.”
ESMA’s survey of securities firms shows how far most are from passing that test. 87% of reported AI use cases are internal-only, and only 17% of firms report complete understanding of AI at board level (ESMA, 2026). The rulebook will wait. The question is already on the table, and most firms cannot yet answer it.
The Obligation That Landed on Time
If the supervisors ask the question inside the firm, the law now asks it at the front door. From 2 August, Article 50 requires firms deploying AI systems to tell people when they are interacting with one, unless that is obvious to a reasonably well-informed person, while providers must mark generated content so machines can detect it (Addleshaw Goddard, 2026). The Commission’s July guidelines add that the notice must be clear and distinguishable at first contact. A line in the terms and conditions does not count.
Notice what survived the delay, too. The technical rules on how a high-risk model is built can wait until 2027, but the rule about what a customer is told cannot. The supervisors are testing whether firms can explain their AI; the law is testing whether firms will admit to it. Both are questions of words before they are questions of code.
The awkward part is that boards are answering a different question in public. Through the spring, HSBC, Standard Chartered and Nordea talked openly about roles that AI would remove, while the European Banking Authority was meeting national supervisors to press for human oversight of AI in credit decisions (Bloomberg, 2026). One story is told to investors, another to regulators, and the customer hears neither; this silence is the gap.
Who Is Putting Their AI Into Words
So who is answering out loud? Some 65% of firms in ESMA’s survey name human oversight as their main safeguard, yet score their own explainability at 3.3 out of 5, lower still for generative models. Firms know what they are supposed to say about their AI, but fewer can show it at present.
The ones getting ahead have started by writing it down. When Revolut launched its AIR assistant to UK customers in April, the data position went out with the product: personal information is not stored by third-party AI partners or used to train external models (Tech.eu, 2026). The explanation shipped on day one, not after the first complaint.
None of this is marketing in the usual sense. It is documentation: what the model does, where a person sits in the loop, what a customer is told and when. It reads plainly because a supervisor, a customer and a journalist will all read it, and none of them trusts the glossy version. Wherever you sit, the wording is the evidence. The firms that will look ready in December 2027 are the ones whose AI can be read about today.
What the Delay Really Bought
The thought to take from the AI Act’s postponement is this: explanation has quietly become a form of control, and control is now shown in words. The supervisor asks whether you can explain the model; the law asks whether you told the customer it was there; and the customer increasingly asks an AI assistant of their own, which answers from whatever you have published. Three readers, one text.
That makes the next sixteen months more a writing assignment than a reprieve. Firms that use the time to put their AI into plain language, what it does, who checks it, what a customer is told, will have their compliance evidence, customer communication and reputation drafted in one pass. Firms that wait for the technical standards will have documentation nobody outside the model risk team can read. The deadline moved to 2027, but the explaining can start this afternoon. The firms that do it plainly will be the easiest to trust when the rules catch up, and the easiest for any reader, human or machine, to describe with confidence.
Ready to be the answer?
When a prospect asks an AI assistant who to trust with their money, their compliance or their marketing, the reply is drawn from what you have published. Contentifai helps B2B firms make their expertise clear, credible and readable by humans and machines alike, starting with an AI Readiness Check of your website. See how your firm reads at contentifai.agency.
Jeremy Rodgers is Creative Director at Contentifai, a UK content marketing agency for B2B firms. Contentifai pairs human writers with AI-supported workflows, helping brands stay visible as search gives way to AI answers.
