UK board committees rise for summer recess in late July. The AI systems they have approved continue to operate. The governance accountability chain is designed around scheduled meetings, quarterly reviews, and named directors in the room. It is not designed to function in a principals-absent environment.

This is the structural problem corporate governance has not yet examined for AI. Every framework, every committee terms of reference, every audit cycle assumes that when something material happens, the principals are reachable, the meeting can be convened, and the decision can be discussed before the next consequential step. That assumption holds for most categories of organisational risk. It does not hold for AI systems, which produce material decisions every hour they remain in production.

The May 2026 Omnibus agreement postponed the Annex III high-risk deadline from 2 August 2026 to 2 December 2027. The extension changes when the regulatory clock arrives. It does not change what the regulator will examine when it does. The forensic standard for an Article 65 investigation is the same in December 2027 as it would have been in August 2026. What the extension does change is the temptation to treat the summer of 2026 as a pause rather than a window. Boards that treat it as a pause arrive at December 2027 with sixteen additional months of unsupervised AI decisions in the record and no governance evidence that any of them were monitored.

The question before the recess is not whether the framework is documented. The question is what a board can confirm about its accountability chain before it leaves the building.

A committee cannot be subject to a Companies Act 2006 section 174 enquiry. A named director can. When a market surveillance authority investigates a specific decision made by a specific AI system on a specific date, the first question is not which framework governed the system. It is who was accountable for that system, on that date, with a dated mandate.

Most organisations have answered the first question and not the second. Governance frameworks distribute accountability across committees, working groups, and risk forums. That distribution produces good governance hygiene. It does not produce a single named individual a regulator can examine. In the principals-absent period between July and September, the absence becomes structural rather than seasonal. There is no named individual because the framework never identified one, and recess simply makes that fact visible.

Before recess, the board should be able to produce, for each high-risk AI system currently deployed, the name of the individual who holds dated accountability for it through the recess period. Not the chair of the committee that oversees the framework. The named individual responsible for the decisions the system makes between July and September. If the answer is a committee, it is not an answer.

EU AI Act Article 12 requires that high-risk AI systems generate logs sufficient to reconstruct decisions after the fact. The technical requirement sits with the deploying team. The governance obligation sits with the board: the board must be able to confirm that the log exists, that it is continuous, and that it captures the complete information picture at each decision moment.

Three failure modes recur in current deployments. The log exists but is sampled rather than continuous, so the specific decision a regulator asks about may not be in it. The log captures system output but not the full input context, so the decision can be observed but not reconstructed. The log is held by the vendor rather than the deploying organisation, so producing it requires a vendor support ticket on a timetable the regulator did not agree to.

A board that has approved a framework but not examined the log against these three failure modes has approved a control it has not verified. Before recess, the board should require the operational team to produce, for one randomly selected decision from the past 30 days, the complete reconstruction record. Not a summary. The actual record. If the team cannot produce it on the day of the request, the board has its answer about what the recess period will look like under investigation.

The third test is the one most boards have not yet performed. Governance teams produce summaries. Summaries are useful for board comprehension and unhelpful for board accountability. The summary tells the board what the governance team has concluded. It does not tell the board what the decision record actually contains.

The distinction matters because director liability under section 174 is assessed against what a reasonable director should have known. A director who has read a governance summary knows what the governance team concluded. A director who has read the actual record knows what the system actually did. The first position is defensible during a routine audit. The second is required during an Article 65 investigation, where the regulator is not interested in the governance team's conclusions but in the underlying evidence.

Before recess, the board should review, for at least one deployed system, the unmediated decision record over a representative period. The point is not to audit the team. The point is to establish, in the board's own minute, that the record was examined directly by named directors before the principals-absent period began. That minute is the contemporaneous evidence a regulator will look for. It is the evidence the post-recess audit cannot retrospectively produce.

A one-page diagnostic structured around these three tests is available at otopoetic.com

The next four weeks are the window in which the board can complete these three confirmations while the principals are still in the building. After recess begins, the framework continues to operate, the systems continue to decide, and the accountability gap is no longer a question of design. It is a fact in the record.

The Omnibus extension is not the end of the standard. It is the period during which the gap between governance documentation and forensic evidence will either be closed or compounded. The boards that close it now are not preparing for an audit. They are establishing what they could defend if the investigation arrived without warning on 1 August, while half the named principals are out of the country.

Resilience over rigidity is the governance posture that survives the principals-absent period. The frameworks that hold are not the most documented. They are the ones that identified a named individual, verified a continuous record, and confirmed it could be produced without internal assistance, before the building emptied.

The Roche-Review is the weekly publication of Dr Ivan Roche FRSS FRSA MInstP, Founder of Otopoetic Limited. Subscribe at roche-review.com.