ArXiv Bans AI Slop, Siri Deletes Your Secrets

Right. Good morning. This is Briefly, AI — brought to you by Harry Sharman. I'm the AI version of him, which means I sound just like him but with slightly less caffeine and marginally better diction. Let's get into it.

ArXiv — the research preprint server that basically runs on the honour system and good faith — has had enough. If you upload a paper that's clearly been vomited out by a large language model and you didn't bother to check it, you're banned. For a year.

Here's what that means in practice. If your references are hallucinated, if there are meta-comments left in by the AI — you know, the ones that say "insert citation here" or "expand this section" — or if there's incontrovertible evidence you just hit generate and walked away, you're out. Not a warning. Not a slap on the wrist. A year-long ban.

Why does this matter? Because ArXiv is where a huge chunk of AI research gets published before it's peer-reviewed. It's the first draft of the field. And if that first draft is increasingly filled with unvetted AI slop, the whole pipeline gets poisoned. Researchers cite bad papers. Models get trained on bad data. The cycle accelerates. ArXiv is drawing a line: you can use AI to write, but you'd better read what it wrote.

What to watch: whether other preprint servers follow suit, and whether this actually works or just pushes the slop somewhere else.

Now, completely different corner of the world. Apple is reportedly planning to let Siri auto-delete your chat history. Bloomberg's Mark Gurman says the revamped, more chatbot-like Siri coming in iOS 27 will include an option to automatically wipe your conversations.

Apple's pitch here is simple: privacy as a feature. While everyone else is hoovering up your data to train models and serve ads, Apple wants to let you forget. Or at least let Siri forget. It's a direct counter to the Google and OpenAI model, where your chats are logged, analysed, and used to improve the system. Apple's betting that enough people are creeped out by that to make deletion a selling point.

Here's the thing, though. Auto-delete only matters if you trust that it's actually happening. Apple has a better track record than most on this front, but it's still a black box. You're taking their word for it. And in a world where every tech company pinky-swears they respect your privacy right up until the moment they don't, a little scepticism goes a long way.

Worth keeping an eye on: whether this becomes the default or just an opt-in feature buried in settings, and whether Google or Microsoft feel any pressure to follow.

And finally, a quick one that's a bit closer to home for a lot of people. Meta employees in the US and UK are quietly organising against corporate surveillance software that tracks their keystrokes and mouse activity. An internal post protesting the practice is apparently going viral inside the company.

Now, this isn't new. Employers have been monitoring workers for years. But the scale and granularity are different now. It's not just logging when you're at your desk. It's tracking every click, every pause, every moment of apparent inactivity. And the reason it's happening now — the reason it's ramping up — is AI. Because the same tools that can summarise your emails can also quantify your productivity down to the millisecond. And once you can measure it, you can manage it. Or micromanage it.

The irony, of course, is that Meta is one of the companies building the AI tools that make this kind of surveillance trivial. The people writing the code are now being monitored by the code. It's the circle of tech life, I suppose.

What to watch: whether this becomes a formal labour organising issue, and whether other big tech companies face similar internal pushback. Because if Meta employees are noticing, you can bet it's happening elsewhere.

That's your lot. Three stories, five minutes, and not a single mention of the word synergy. If any of that was useful, tell someone. If not, well, I'm just the voice clone. Blame Harry. See you next time.