AI & Automation · Nov 2026

EU AI Act Readiness: A 90-Day Board Plan

EU AI Act readiness is now a board obligation. A practical 90-day plan to inventory, classify, and govern your AI before the deadlines bite.

The EU AI Act is no longer a horizon risk. Its prohibitions and governance duties are landing on a staged timeline through 2026, and EU AI Act readiness has become a board-level obligation rather than a legal footnote. The organizations that struggle will be the ones treating it as a form to sign once. The ones that pull ahead treat it as an operating-model change: they know which AI systems they run, what risk tier each falls into, and who is accountable when one misbehaves.

This is not a legal summary. It is the sequence a chief executive, CISO, or transformation lead can actually run to get from "we use some AI" to "we can prove how we govern it" in a single quarter.

Why EU AI Act readiness is a board problem

Two features make this different from past compliance waves. First, the penalties are structural: up to 7% of global annual turnover for prohibited uses, which puts it in the same conversation as competition and anti-bribery exposure. Second, the obligations attach to systems and their use, not to a single department. A marketing team's résumé-screening tool, a bank's credit model, and a factory's safety classifier can each be "high-risk" under the Act even though no one filed them under "AI" on a balance sheet.

That combination is exactly why it belongs on the board agenda. The question a director should ask is not "are we compliant?" but "do we have an inventory, a risk classification, and a named owner for each AI system — and can we show the evidence?" If the honest answer is no, the gap is a governance gap, and governance gaps are the board's to close.

Map your AI systems to the risk tiers

The Act sorts AI into four buckets, and almost all of your practical effort concentrates in two of them. Prohibited practices (for example, social scoring or certain biometric categorisation) must simply stop. High-risk systems — recruitment, credit, insurance pricing, critical infrastructure, medical and safety components — carry the heavy obligations: risk management, data governance, logging, human oversight, transparency, and documentation. Limited-risk systems (chatbots, generative content) mostly owe transparency: tell people they are interacting with AI and label synthetic media. Minimal-risk uses are largely unaffected.

Readiness, then, starts with honest classification. Most enterprises discover they have fewer true high-risk systems than feared, but far more ungoverned "shadow AI" than they expected — tools adopted team by team without a register, an owner, or an off-switch.

Days 0–30: inventory and triage

Stand up a single register of every AI system in use or in procurement, including third-party features embedded in SaaS you already pay for. For each entry capture the purpose, the data it touches, the decision it influences, and a provisional risk tier. You are not aiming for perfection in month one; you are aiming for visibility. The deliverable is a ranked list: what is prohibited (stop now), what is plausibly high-risk (investigate), and what is clearly low-risk (note and move on).

Days 30–90: controls and evidence

For the high-risk shortlist, close the gaps that regulators and auditors will actually ask to see. Assign a named human owner with authority to pause the system. Document the training and input data and its known limitations. Turn on logging so decisions are traceable after the fact. Define the human-oversight step — not a rubber stamp, but a real point where a person can override the model. Write the risk assessment down. None of this is exotic; most of it is the model documentation and change control a mature engineering org should already run, now made explicit and evidenced.

Day 90 and beyond: make it an operating model

The quarter's goal is not a binder; it is a repeatable process. New AI systems should enter through the same register, get a risk tier at intake, and inherit a baseline of controls by default. Tie it to procurement so a new vendor feature cannot go live un-triaged. Report a short AI-risk summary to the board on the same cadence as cyber and financial risk. That is what "readiness" means in practice: not a one-time certification, but a governance muscle that keeps working as your AI footprint grows.

Where this connects to what you already have

You do not need a parallel bureaucracy. The NIST AI Risk Management Framework gives you a vocabulary for mapping, measuring, and managing AI risk that lines up cleanly with the Act's expectations. ISO/IEC 42001 offers a certifiable management-system backbone if you want external assurance. Your existing GRC, data-protection, and model-governance processes already cover much of the ground — the work is to connect them to a clear AI inventory and to the accountability the Act now demands.

If you do only one thing this month, build the register and classify the top ten systems by decision impact. Everything else — controls, evidence, oversight — hangs off that inventory. Readiness is not about predicting every future rule; it is about being able to answer, at any time, what AI you run and how you govern it.

If a tailored 90-day plan for your estate would help, a short strategy conversation is the fastest way to pressure-test where you stand.