Update, 8 August 2026. Since 2 August, the Article 50 transparency obligations discussed below are in force. The evidential test set out here now applies in the present tense; the underlying argument is unchanged.
Most firms are treating Article 50 of the EU AI Act as a technical compliance task. For a UK financial services firm under the Senior Managers regime, it is not. Whichever senior manager owns a decision area already owns the Article 50 obligations that attach to it, and the Statement of Responsibilities that says so has been filed with the FCA.
Since 2 August the deployer obligations have applied. Firms must disclose when a person is interacting with an AI system, label deepfakes, and flag AI-generated content published on matters of public interest. Preparation has mostly been run as an inventory of systems in scope, a review of disclosure wording and a workstream in the second line. That misses the sharper half of the exposure. The Senior Managers and Certification Regime requires accountability for each area of regulated activity to be mapped to a named senior manager, recorded and filed, and it does not distinguish between a decision made by a human and one materially influenced by an AI system.
The two regimes therefore converge on a single evidential test, reached from two directions.
The EU regulation asks who is accountable for the deployer's obligations. The domestic regime has already required the answer to be written down, attached to a person, and filed. What a regulator can now ask is narrow: can the named senior manager produce a dated record showing they reviewed the obligation before it applied?
Figure: where the two regimes meet
The failure mode is specific. A firm maps Article 50 readiness to a technical function. Its Statement of Responsibilities names a senior manager. The work is done in one place; the accountability is filed in another. If the named individual holds no dated record of review, the firm carries two accountability gaps that do not agree with each other, and both are visible from documents the regulator already holds.
Closing the gap
Closing that gap is not a documentation exercise after the fact. It requires knowing, for each material AI-assisted decision area, who is named, what they reviewed, when, and whether the date on that record would survive being challenged.
A filed responsibility without a dated review is a promise nobody has checked.
Governance of this kind is a standing rhythm of review, not a pre-deadline scramble: each named owner re-confirms what they reviewed, and the date moves with the change.
For your last material AI-assisted decision, which named senior manager could produce a dated record showing they reviewed the obligation before it applied?