One minister can decide what counts as a person being involved in a decision about you. Six months on, that power has not been used.

Every significant automated decision made about a British person now turns on four words: meaningful human involvement. If a decision has it, the new rules barely apply. If it does not, a set of safeguards switches on.

The Act does not say what the phrase means.

Article 22D hands the Secretary of State power to make regulations defining meaningful human involvement, and to define what counts as a significant decision, and to prescribe further safeguards. The explanatory notes say the term may be clarified by secondary legislation in light of constantly new technologies. Six months after the rules took effect, no such regulations have been made.

That is not a scandal and it is not one party's failure. It is worth saying plainly that the power is recent, that the department has a queue, and that a rushed definition of a phrase this load-bearing would be worse than a slow one. Anybody demanding it be done by Friday has not thought about what a bad definition would fix in place.

But three things follow from the gap and they are all live now.

The first is that the boundary can move without Parliament. Article 22D lets the definition be set, and reset, by regulation. Whichever way you think the balance should fall, the mechanism means the reach of the entire regime can change without a vote.

The second is that the vacuum is being filled anyway. The Information Commissioner's Office published draft guidance on 31 March and consulted on it until 29 May, with a statutory code still to come. The draft takes a strict line. Review must happen before the decision is applied, and every time, not as occasional spot checks. Building the system does not count as being involved in its decisions. A reviewer who holds the authority but never overturns anything is not involved either. That is a regulator's reading of a term the law has not defined, and firms are building against it because it is the only steer available.

The third is the one this strand exists for. In the ICO's own recruitment work, employers described their tools as decision support, while the regulator's evidence suggested there was often no meaningful human involvement and the tools were producing decisions with significant effects on candidates. The gap between what firms believe they are doing and what they are doing is exactly the gap a definition would close.

None of which is to say the department is idle. The government is developing the secondary legislation the ICO needs for its statutory code on AI and automated decisions, and the ICO says that work is under way. A separate government call for evidence, on whether data regulation helps or hinders AI, closes on 9 September. Neither of those is a definition of meaningful human involvement, and neither uses the Article 22D power, but anyone telling you nothing is happening is wrong.

What we could not find is the narrower thing. We looked for a Commons or Lords debate, a select committee session, or a ministerial statement addressing what the phrase itself should mean, and did not find one. That is what we found, not proof there is nothing.

WHY IT MATTERS HERE: The question of which office jobs survive contact with a machine is, in this sector, a question about four undefined words. The person who gets to define them has not yet done so, and the definition can be changed again later by the same route.