The Metropolitan Police spent the past fortnight investigating hundreds of its own officers, after a Palantir analytics layer it had bolted on to its existing workforce-monitoring software started flagging breaches within days of going live. Within seven days of the pilot, the Professional Standards directorate had three arrests on suspicion of fraud and serious misconduct, 98 officers under investigation for abuse of the shift-roster IT system, 500 prevention notices, 42 senior leaders being assessed for hybrid-working breaches, and a dozen officers facing gross misconduct over undeclared Freemasonry links.

Sir Mark Rowley, the Commissioner, called it identifying risk earlier, acting faster, and being fairer and more consistent. The Police Federation, which represents around 30,000 Met officers, called it “an outrageous and unforgivable invasion of privacy” and is now advising its members to leave Met-issued devices at home when they’re off duty.

The Federation is preparing a legal challenge under Article 8 of the Human Rights Act and GDPR. The officers it represents weren’t told the Lawful Business Monitoring software they signed up to now ran a US firm’s analytics on top. And all of this is happening in the same week that 229,000 members of the public signed petitions demanding the UK government tear up Palantir’s roughly £600m portfolio of contracts across the NHS, the Ministry of Defence, local councils, and policing.

Underneath the headline is a bigger story. Workplace AI surveillance has crossed out of the commercial world and into the place where every taxpayer is the customer — and the political ceiling that’s now forming around state surveillance is going to set the practical ceiling for what private employers can deploy too. If you’re running a UK business and thinking about workplace AI, the Met case is the leading indicator you need to read.

Why this is different from Meta logging keystrokes

We’ve spent the past two years watching this kind of tooling roll out across the private sector — Meta logging keystrokes, Amazon timing toilet breaks, Microsoft scoring productivity inside Teams. The reaction was always somewhere between resignation and weary distaste. You could leave the job. You couldn’t really vote against it.

The Met case is in a different category, and it’s worth being precise about why.

Start with who pays for it. The principal isn’t an employer; it’s the public. The Met’s payroll is funded by Londoners through council tax, the Home Office settlement, and a stack of grants that ultimately comes from general taxation. When a private firm decides to surveil its own staff, the social contract is, broadly, “you took the salary, you knew the deal.” When the state surveils its own staff, the social contract is “you elected the people who signed off on this technology, and you’re paying for it.” Those aren’t the same.

The people being watched also have organised resistance built into the role. A 30,000-strong federation with the right to threaten judicial review isn’t an HR forum. It can pull a procurement decision into the High Court and force a public test of whether the technology is even lawful.

There’s a fourth thing about the technology itself worth being honest about. Tools like this are pattern-finders, not judgment engines. They notice the shift-roster anomaly, the device-location signature, the absence-attendance correlation. They don’t notice whether the officer was caring for a sick parent, navigating a medical appointment, or recovering from a critical incident the previous night. The system’s job is to flag; a human’s job is to read context, intent, and fairness back into the case. That’s a workable division of labour right up until the moment detection scales faster than the people doing the reading.

And every flagged officer is now also a story. A private employer can move discreetly through a misconduct process. A police force can’t — every arrest, every gross-misconduct hearing, every undeclared-Freemasonry case is on the front of The Times the next morning, and the news cycle scales right along with the detection.

You can see why this arrived first as commentary and is becoming a legitimacy crisis. I’m not sure anyone — the Met, the Federation, the departments watching anxiously from the sidelines — actually knows yet how this lands in court or at the next election. The experiment is running in public, with everyone in the room.

The procurement question nobody is asking

There’s a more boring story underneath the political one, and it matters more than it sounds. The Met’s procurement of the Palantir layer reportedly came in under the £500,000 threshold that triggers full competitive tendering. The Federation’s lawyers are now arguing — reasonably — that a threshold designed for desk chairs and printer toner isn’t the right instrument for buying surveillance-grade AI.

This is becoming infrastructure, not a one-off pilot. BlueLight Commercial, the joint Home Office and policing buying body, is currently looking for a lead delivery partner to build “Police.AI” — a national platform for testing and deploying AI tools across the 43 forces. The formal tender is expected in early June, with a two-year contract starting in September. That platform will, in time, become the front door through which most British police AI procurement runs. Once a tool is sitting on infrastructure, the political fight is no longer about one force or one vendor; it’s about the operating model.

It’s also worth saying out loud that Palantir isn’t in the public sector by accident. The company’s core product is a pattern-of-life engine built originally for intelligence work, and what the petition is reacting to — even if it doesn’t put it this way — is the migration of a commercial surveillance product into the operating fabric of British public services. That migration matters whether or not the underlying technology works.

Right now, nobody in Parliament has had to vote on what the threshold should be for buying analytics that can track an officer’s location around the clock, score their behaviour, and trigger a misconduct investigation. Procurement convention will set it instead. And as I’ve argued in another piece on the UK’s broader AI policy direction, the gap between what the government says it’s doing and what’s actually getting bought is wider than any of the public statements suggest.

What this means if you’re rolling out AI in your own business

The reflex reading of all this is that the Met case is a public-sector story. It isn’t only that. The political ceiling for what the public will tolerate from public-sector AI sets the practical ceiling for what private employers can get away with too. The same press, the same lawyers, the same campaigning organisations are all watching.

A few practical lessons fall out of this if you’re inside a business considering an AI deployment with any monitoring component:

  • Tell people what you’ve bought. The single most damaging fact in the Met story isn’t the analytics layer. It’s that officers weren’t told the layer had been added to software they were already using. “Lawful Business Monitoring” became something materially different overnight, and no consent was renewed.
  • Treat surveillance-grade AI as a different procurement category. A tool that can track location, score behaviour, and trigger discipline isn’t in the same league as a CRM bolt-on. Buy it, govern it, and disclose it on those terms. The £500k procurement floor was designed for office furniture; it shouldn’t be the floor for a system that produces dismissal evidence. The companies that get this right will look more like the governance practice I’ve sketched out before — boring, formal, and signed off in advance — and less like a lunchtime IT decision.
  • Build the appeal route before the model. When the Met’s tool flagged 500 prevention notices in seven days, it created 500 conversations the Met didn’t have time to have well. AI scales detection faster than due process can keep up with it. If the appeal route isn’t faster than the model, you’ve got a fairness problem and a courtroom problem on the same day.
  • Watch where your supplier is also working. Procurement teams that haven’t started asking those questions are about to.
These aren’t far-future concerns. The market for workplace AI is moving from “what could we measure?” to “what should we measure, and who do we have to tell?” The businesses getting that order right will end up more credible than the ones racing the technology in first and the disclosure later.

The wider point

The pilot is now being tested in two places at once: in court, by the Police Federation, and at the ballot box, by the 229,000 people who’ve already signed petitions. Both tests are about the same thing — whether a tool that produces convincing-looking patterns about people is allowed to drive consequential decisions about them without anyone having properly agreed it should.

The Met’s Palantir tool will probably keep working. Whether it keeps being allowed to work is no longer a question the Met can answer in a press conference on its own; the Federation, the courts, and the 229,000 petition signatories all hold a piece of that answer too — and the worst-case outcomes here are concrete, not abstract. A judicial review that strikes the deployment down. A procurement freeze across the wider Police.AI programme. A senior dismissal that becomes a precedent. The same widening of the audience is on its way to every business rolling out workplace AI, and the ones that handle it best will be the ones who act now as if the public has already arrived in the room.