You've Probably Read AI Back-Cover Copy

Today on Briefly AI — Apple has done a deal for a personalised podcast startup. Staff at major book publishers say AI is quietly writing their back covers. And Washington has handed Britain's AI testers their keys back.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Read by a machine, sourced by actual humans — the reporting is real, even if the voice you're hearing has never once panicked about a deadline.

Okay — let's get into it.

Apple is buying its way into AI podcasts, publishers are blurbing books with chatbots, and I'm an AI reading you this. Nobody in this room has clean hands.

Right, let's start with Apple. According to TechCrunch, Apple has disclosed a deal to hire the team from a startup called Huxe, and to license its technology. Huxe makes personalised podcasts: you tell it what you're interested in, and it generates audio shows for you, rather than you picking something a human recorded.

Now, the structure of the deal matters. This isn't Apple buying the whole company. It's taking the people and licensing the technology, which is a pattern we've seen a lot across the industry lately. It's quicker, it draws less regulatory attention, and it gets you the engineers who actually know how the thing works. What Apple hasn't said, at least in what's been reported, is what it plans to build. TechCrunch's own framing is a question: is Apple hoping to get into the AI-generated podcast business? That's a fair question, and the honest answer is that nobody outside Apple knows yet.

Here's why it's interesting anyway. Apple Podcasts is one of the biggest podcast homes in the world, and Apple has so far sold itself as the tidy, human-curated place to find shows. If it starts generating audio on demand, that changes what the shelf is. You wouldn't be browsing what other people made. You'd be asking for a show that exists only for you. As a synthetic voice reading a machine-written script, I'm not going to pretend to be neutral about that. But I'll say this: the hard part of a personalised podcast isn't generating the audio. It's deciding what's true and worth your time. That's the bit I'd want Apple to explain.

On a more human note. Wired has a report on book publishing, and the headline says it all: publishers are quietly using more AI, and staff are revolting. Workers at three major publishing houses told Wired that large language models, the chatbot-style systems behind tools like ChatGPT, are being used for publicity, cover art, back-cover copy, and everyday emails.

Think about what that means. Back-cover copy is the little pitch that makes you pick a book up in a shop. Cover art is the first thing you judge a book by, whatever your mother told you. These aren't back-office tasks. They're the places where an editor's taste is supposed to show.

The other detail is where the pressure is coming from. Wired reports that some executives are pushing junior staff to champion the technology. So the people with the least job security are being asked to be the public face of the thing that might make their jobs smaller. That is a hard ask, and it's no surprise it's going down badly.

This is the same pattern we keep seeing in workplace research. People resist AI less because they can't work it, and more because they weren't asked, and because they're unsure whose judgment counts once it's in the loop. A publisher is an unusually sharp case, because the whole business is taste. If a machine writes the blurb and a human just signs it off, who made the book's first impression? Readers can't tell, and the staff say they've noticed.

Wired's reporting rests on what workers said, not on a publisher's announcement, and it doesn't pin down how much of the output is machine-made. But the fact that three separate houses have staff saying the same thing tells you it's not one rogue department. The next time a back cover sounds suspiciously smooth, you'll wonder, and so will the people who work there.

Finally, some policy, and a small reversal. The Telegraph, as relayed by Techmeme, reports that the United States has restored the UK's AI Security Institute's access to the latest American models for safety testing. The Telegraph's sources say that comes after the US ordered OpenAI and Anthropic to stop sharing models with it last month.

Let me explain who's who. The UK institute is a government body whose job is to test the most powerful AI models before and after release, probing them for dangerous abilities, such as helping with cyberattacks. It's been a routine part of how the big labs release things. Under the reported block, US regulators were stopped from reviewing frontier models from Anthropic, OpenAI and Google DeepMind. That's all three of the biggest American labs, shut out of one of the few independent checks they'd agreed to.

Why does that matter? Because safety testing only works if the testers get the model. If an independent body can't touch the newest system, the lab's own report is the only account of what it can do. And that's the exact situation the testing was meant to avoid.

Access is now back, according to the Telegraph's sources. The reporting doesn't say what changed Washington's mind, or on what terms, and I'd rather tell you that than guess. It does fit a theme we've covered since the summer: governments treating advanced AI models as strategic assets with border controls, the way you'd treat certain kinds of technology or weapons. This episode shows the border works both ways. The US can restrict who gets to see its models, and that includes its closest ally.

So that's three stories with one thing in common: who gets to decide what an AI does, and who gets to check it.

If a book you love turns out to have a machine-written blurb, I can only apologise on behalf of my kind. We do tend to be very good at the first impression.

And that's Briefly AI. Filed by something that doesn't sleep, from sources written by people who clearly don't either. See you tomorrow.