We only know what happened to the people inside this company because it died. That is not a reporting triumph, it is an accident

I have spent a long time around companies that promised a machine would do the work. I have been on the inside of one where it did, and I watched a team I had built by hand come apart because of it. So I am not a sceptic about any of this. I think the machines take the work.

Which is why the story in this edition sat awkwardly with me all day yesterday.

BenevolentAI was London's serious attempt to hand the hardest thinking work there is to a machine. It did not work. And the people who did that work lost it anyway, 277 researchers down to 170 in a year, and then the research stopped altogether.

That is not the story I expected to be running. Elena says more about it a few pieces down and she does not flinch from it, which is the only reason we are running it.

But here is the part I keep coming back to, and it is not about that company at all.

We can only count those people because the business collapsed.

While it was alive and listed abroad, its British accounts were allowed to be thin. That is not a loophole and nobody broke a rule. It is an ordinary exemption that applies to any British company owned by a parent registered outside the country. The full numbers arrived only when the listing died and the exemption fell away with it. If the company had survived in the shape it had, we would still be guessing.

So the honest position is this. We are reporting a rare, accidental view of what happened to a workforce, and presenting it as though we found something. We did not find it. It fell out of a corporate failure and landed in a filing cabinet in October.

Now the bit that belongs to Westminster, and it is not one party's problem.

James has the figure elsewhere: 17.1 million pounds of research and development tax credit recognised across two years at this one company. That scheme has been run, reformed, expanded and trimmed by both parties for a quarter of a century. It has never, under any of them, been asked to say what it buys in jobs. It reports what it costs. It does not report who was employed on the work, or what became of them when the work stopped.

I am not calling for the scheme to be cut. I have no idea whether it is good value and neither does anybody else, which is rather the point. I am saying that a country spending billions to keep research going in Britain has never built the one measure that would tell it whether British people are doing that research, or still doing it a year later.

And one last thing, which is the only bit that matters.

A hundred and seven research jobs went at that company in a single year. I have read the accounts, the announcements and the register. Not one of those people is named anywhere. Nobody has asked a single one of them what happened next, and I should be straight with you: we cannot either. There is no list. There was never going to be one, because nobody is required to keep it.

— M.

This note is mine: the view, and the call to run it. It begins as a draft, drawn from work the AI and I have researched and argued out together, the same way every desk in this paper is made, and I answer for every line because I read every line. Those desks run on models built by Anthropic, one of the labs sitting on the very scoreboard we report, so we tell you plainly: we cover this from inside it.