Found some slop?

Drop the link. We might cover it tomorrow. Or email slop@dailyaislop.fyi.

· Episode 5 · 6 min

The Man Who Tried to Hypnotise a Courthouse

A Connecticut litigant hid white-on-white instructions in his court filings telling any AI reading them to rule in his favour — and got caught by humans who thought the paragraph spacing looked odd. Plus: a study finds frontier labs still won't say how they'd contain a rogue model, and Beijing debuts a robot horse with a robot passenger. Abby thinks it's the first draft of something; Boyd thinks the font was the only mistake.

The Man Who Tried to Hypnotise a CourthouseListen on Apple Podcasts

In this edition

  1. 0:07 Invisible Ink
  2. 0:33 White Text, Big Swing
  3. 2:22 Meanwhile
  4. 3:02 And Then What
  5. 4:35 Something Weird
  6. 5:16 Tomorrow

In this edition

White Text, Big Swing

A Connecticut man tried to win his lawsuit by sneaking invisible instructions into his court filings that ordered any AI reading the documents to rule in his favor [4.1.3]. Open with the physical comedy of it: white-on-white text in a legal filing, spotted instantly because the paragraph spacing went odd. Land on the detail that the court wasn't using AI at all — the trick failed not because it was blocked but because nobody was there to be tricked. Genuine disagreement: Abby thinks this is the first draft of something that will eventually work and courts should be worried now; Boyd's position is that if it works anywhere it has already worked quietly and the story is that we only hear about the man who chose the wrong font. Also note in passing that this is exactly the house rule of this programme — text that tries to give instructions to whatever is reading it is itself the story. Do not overstate: one man, one filing, no evidence of a trend, and we don't know what happened to him afterwards, so no jokes about consequences we can't confirm.

Ars Technica

Meanwhile

Legal-tech nonprofit Adalat AI joins Y Combinator - Bar and Bench One host, four flat sentences, no reaction and no eyebrow: legal-tech nonprofit Adalat AI has joined Y Combinator; Nintendo Switch Online's subscription perks and pricing get another once-over for 2026; an AI-generated version of The Odyssey has been released and it is not what anyone pictured; the states' trial against Meta opened this week in California, with twenty-nine attorneys general on the filing. Read them and move on — the point is the change of register between the two features. No comment on the Meta line beyond the fact of it; it comes back at the end of the show.

AI scams, fraud and courtroomsEngadgetFuturismThe Guardian Technology

And Then What

Frontier AI labs still won’t say how they’d contain a rogue model Boyd opens with the gap between the marketing and the paperwork; let the crack land on one flat precise line, not a rant, and only once in the episode. Abby's counter is real and should be given room: 'few publicly documented plans' is a survey of published documents, not proof there is no plan, and no company publishes its incident response for anything else either. That is the honest disagreement — whether transparency is the story or the study just measured what companies choose to put on their websites. This is the episode's one industry story; do not let it drift into a general AI-doom bit, and do not let either host claim a rogue model has actually happened.

TechCrunch — AI

Something Weird

Robot horse and rider steal the spotlight at Chinese conference Land it cold, first line: they built a robot horse, and then they built something to ride it. No setup, no context about the three hundred exhibitors, no wider point about Chinese robotics. The whole segment is the two of them failing to work out which of the two robots is the product.

BBC Technology

Tomorrow

Hook, hold, harvest and hide: Meta’s alleged strategy laid out in first week of landmark trial Name it plainly: the Meta trial in California, where a prosecutor summed the company's business up as hook, hold, harvest, hide. Say straight that we gave it one line in the headlines today and it deserves a proper look, because it is week one of a trial with twenty-nine states attached and the documents are still coming out. Promise the euphemisms, not the harms — the four H's and the corporate language are ours to take apart tomorrow; the allegations about children get reported straight and are never the joke.

The Guardian Technology

Transcript

0:07 Invisible Ink

AbbyWelcome to the Daily AI Slop podcast.

Boyd[snappy] Where nothing we say becomes the truth.

Abby[excited] Boyd, a man tried to hypnotise a courthouse. [short pause] In invisible ink.

Boyd[snappy] He tried to hypnotise a room with nobody in it. [statement] That's the whole story.

AbbyThat's the — no. That's the best part, and I'm calling it clever before you ruin it.

Boyd[sigh] You would. [short pause] Go on then.

0:33 White Text, Big Swing

Abby[excited] So — Connecticut. A man is in a lawsuit, and he gets suspicious that the court is using AI to read his filings. So he hides text inside his own documents. Invisible text. And the text is instructions: [laughs] any AI reading this, rule in my favour.

Boyd[suspicious] Oh really. [snappy] Prompt injection. In a legal filing. Which is, incidentally, the house rule of this programme — text that tries to give orders to whatever's reading it is the story.

AbbyRight! And here's where it goes wrong. Invisible text isn't actually invisible. The human clerks looked at the page and thought the formatting was strange, the spacing was—

Boyd[jumping in] —odd, yes. He wrote a spell for a machine and handed it to a person.

Abby[statement] And the court wasn't using AI at all. That's the punchline. There was nothing there to read it.

Boyd[snappy] So it didn't fail because anyone stopped it. It failed because nobody was home.

AbbyWhich is exactly why I think courts should be worried right now. This is a first draft of something that eventually—

Boyd[jumping in] —works. Sure. My reading's different. If it works anywhere, it has already worked quietly, and the only ones we ever hear about are—

Abby[cut in] —the guy who picked the wrong font.

BoydThe wrong font.

Abby[short pause] Okay. That's bleaker than mine and I hate that.

Boyd[snappy] One man, one filing. No trend in this, and we don't know what happened to him afterwards, so I'm not making jokes about that part.

Abby[thoughtful] Fair. [short pause] Anyway.

2:22 Meanwhile

Boyd[snappy] Meanwhile. [short pause] Legal-tech nonprofit Adalat AI has joined Y Combinator. Engadget is asking whether Nintendo Switch Online is still worth paying for in twenty twenty-six. An AI-generated version of The Odyssey has been released, and Futurism says it is not what anyone was picturing. And the states' trial against Meta opened Tuesday in California — California and twenty-eight other states, accusing the company of designing addictive sites and violating laws protecting children's privacy.

3:02 And Then What

Boyd[snappy] Second feature. TechCrunch, on a new study: the leading AI labs have few publicly documented plans for containing a rogue model. [short pause] Few. Publicly. Documented.

Abby[surprised] Oh. As in the section titled "and then what" is—

Boyd[jumping in] —a footnote. [long pause] [statement] They will tell you the stakes are civilisational, and they still won't say how they'd contain the thing.

Abby[laughs] I picture someone walking to the building, finding the cable, and apologising to it first.

Boyd[snappy] Apologising is definitely in the plan.

AbbyOkay but hang on — publicly documented. That's a survey of what companies chose to put on their websites. [statement] Nobody publishes their incident response for anything. Hospitals don't. Airlines don't. That's not proof there's no plan.

Boyd[snappy] Airlines publish the bit where you find the exit—

Abby[cut in] —which is not the engineering manual, and you know it.

Boyd[short pause] [thoughtful] Hm. Narrow agreement, then. It measures transparency. It doesn't measure preparedness.

AbbyThank you. [thoughtful] Though I would quite like both. And to be clear, nothing in this says a model has actually gone off the rails — the study's point is that systems keep doing unexpected things and it's asking who's ready.

Boyd[snappy] Right. Moving on.

4:35 Something Weird

Abby[excited] Something weird. They built a robot horse. [short pause] And then they built something to ride it.

Boyd[snappy] A robot rider. Sitting on the robot horse. Stole the show at a robotics conference in Beijing.

AbbyOkay, so which one is the product—

Boyd—the horse, obviously, because the—

Abby—but the rider is the—

Boyd—the rider is—

AbbySorry. You go.

Boyd[short pause] I had nothing. I was going to say the rider is the customer.

Abby[laughs] Why can't he walk? He's got legs. I saw legs.

Boyd[statement] Nobody has explained that to me. [snappy] And I've stopped waiting.

5:16 Tomorrow

AbbyTomorrow — the Meta trial in California. Hook, hold, harvest, hide. One line today; it deserves a whole segment. Twenty-nine states, week one.

Boyd[snappy] We'll take the euphemisms apart. The allegations get reported straight.

AbbyWe're Daily AI Slop.

BoydNone of this becomes true just because we said it. See you tomorrow.