You Funded Your Own Surveillance Camera

Today on Briefly AI — Texas's governor has frozen funding for AI surveillance cameras that drivers unknowingly paid for themselves through their own insurance bills. Caterpillar is taking decades of lessons from driverless mining trucks and applying them to everyday AI rollouts. And a new paper claims AI can already out-diagnose your doctor — and doctors are not thrilled about it.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. An AI reading you the AI news. The novelty of that sentence is wearing off faster than any of us expected, which is rather the point.

That's the shape of the day. Let's dig in.

Right, quick thought experiment for you. A machine that's hauled rock through the outback for twenty years without a driver, and a camera on a pole that photographs your number plate on the school run — today, weirdly, those are basically the same story. Also, your doctor's having a bit of a moment, and it's not the good kind.

First up, the one your wallet's quietly been financing without you noticing. There's a small camera on a pole you've probably driven past a hundred times without registering it — that's Flock Safety, the AI license-plate reading system now installed by police departments right across the United States, marketed as a way of tracking vehicles to solve crime. And in Texas, it turns out, a huge chunk of that network was paid for by drivers themselves.

According to The Verge, a Texas Tribune investigation found the state has spent more than thirty million dollars on Flock cameras — and most of that came from a one-dollar fee quietly tacked onto every Texan's car insurance policy. Nobody voted on the surveillance line item. It just showed up, a dollar at a time, folded into a bill people were already paying for something else entirely.

Governor Greg Abbott has now frozen state funding for any more of these cameras, and the timing is not subtle — the freeze landed just ahead of the Tribune's story going public, which is the kind of scheduling that tends to happen when someone in government reads a draft before the rest of us do. This matters beyond Texas because Flock's cameras have already been at the centre of national rows over who gets to search the footage — reports of local police feeds being accessed by federal immigration agents, or used to trace people's movements well past whatever crime the camera was installed to catch. Texas is one of the largest state-level deployments in the country, so a freeze here is a real dent in the network, not a footnote.

What we know for certain: the spending happened, the fee funded it, and the freeze is real. What's still open is whether Texas claws back any of that thirty million, or whether other states start asking who's actually paying for the cameras watching them. For now, the practical takeaway is simple — next time you spot one of those little poles by the road, there's a decent chance you helped buy it.

Second story is a nice contrast to a year of headlines about AI agents going off the rails, because it's about a company that's been running autonomous machines with nobody in the driver's seat since long before "AI agent" was a phrase anyone used. That's Caterpillar, and according to TechCrunch, it's now applying what it learned automating mining to the much broader job of rolling out AI across the business.

Some context on why Caterpillar gets to have an opinion here. Its Cat Command system has run driverless haul trucks at remote mine sites — places like the Pilbara in Western Australia and copper mines in Chile — for well over a decade, hauling rock around the clock through dust storms and desert heat, often hundreds of miles from the nearest town. Those trucks have shifted well over a billion tonnes of material with nobody on board. That's about as unforgiving a proving ground for autonomy as exists on Earth — get the software wrong out there, and the consequence isn't a bad review, it's a multi-tonne truck somewhere it shouldn't be.

So what's the actual lesson being carried across? By TechCrunch's account, it's less about the flashy model and more about the unglamorous stuff that makes autonomy survivable: phased rollouts inside tightly controlled areas before anything goes wide, redundant safety systems that assume the AI will occasionally get something wrong, and humans who don't vanish from the loop — they just move to a control room and watch whole fleets instead of one truck. It's a playbook built for machinery that can genuinely hurt someone if the software has a bad day, which makes it a surprisingly good stress test for the far lower-stakes AI agents most office workers are being handed right now.

The result, as reported: Caterpillar is treating its mining playbook as a template for its own enterprise AI deployment — proof, if nothing else, that the company with the least glamorous AI story might have the most battle-tested one. Given how many businesses are currently bolting AI agents onto systems with no real safety net, borrowing a manual written by people used to machines that weigh two hundred tonnes doesn't seem like the worst idea going.

Last one's a proper professional-identity gut-punch, and it's aimed at people we're trained to trust with our lives. Wired reported this week on a new paper arguing that AI is, in a lot of measurable ways, already a better diagnostician than the average doctor — and, unsurprisingly, doctors are not popping champagne.

This isn't coming from nowhere. Back in 2024, Google published research on a conversational diagnostic AI called AMIE, which beat primary care physicians on diagnostic accuracy in blinded case comparisons — and patients in that study actually rated the AI's bedside manner as more empathetic than the humans', which is its own kind of uncomfortable. The paper Wired's covering this week pushes the argument further: across a set of test cases, the AI didn't just match human performance, it out-diagnosed doctors more consistently, especially on the messy, rare, or ambiguous cases where a GP has maybe ninety seconds and a waiting room full of other patients to get to.

Here's the bit that actually matters if you're not a doctor. Diagnosis is one of medicine's most pattern-matching-heavy tasks — feed a model enough symptoms, histories, and outcomes, and it gets very good at spotting the needle in the haystack a tired human might miss on a Friday afternoon. But diagnosis isn't the whole job. It's not the physical exam, it's not the twenty years of trust built with a frightened patient, and it's not who gets sued when the call is wrong. Doctors pushing back on this paper aren't wrong to feel got at — the real argument underneath it is which part of the job an AI gets to take, and which part stays human because someone has to be accountable for it.

What's actually happened so far: the paper is out, it's added fuel to an argument already running hot in medicine, and doctors' professional bodies are, per Wired, distinctly unthrilled about the framing. Nobody's been replaced. But the case for why the person you've seen for fifteen years still has the edge over an algorithm just got a lot harder to make.

So there we are — a mining company quietly teaching the AI industry how not to kill anyone, doctors bracing for a very awkward performance review, and a state government discovering its own citizens crowdfunded a surveillance network one dollar at a time. If your GP seems a bit tense this week, maybe don't bring up the robot.

You can find more at harrysharman.com. Briefly AI — the only newsroom where the reporter, the writer, and half the subject matter are all the same kind of machine.