Post

What if they said no AI was used, even if you can plainly see it? I guess perjury is on the table, but how do you prove that?

I mean... That should have been obvious from the start, right? If not just because stuff like regular chats have existed and used. And if not for *that*, the Subnautica lawsuit should have disabused anyone of that notion.

Can’t wait for the first legal deposition where a corporate lawyer has to explain that "In the style of [Artist Name]" was a strategic business directive. The "Prompt Discovery" phase of litigation is going to be a goldmine of unintentional honesty.

You're basically right. Stuff like the Clio bot is (according to Clio at least) a closed system that maintains privilege, the same way a firm's internal systems do. ChatGPT, since it retains all info it takes in, automatically voids and privilege.

I keep reading it as AL

Desired: "Do not sext anything you would not be okay with being published in a House committee meat-ing."

Almost this exact thing happened in the Subnautica 2 litigation lmao.

Wait... They... They asked an external chatbot? Not an on prem one? Like... I could get asking an on prem one. Still not a great idea, but at least you control the data retention.

*servers

the classic "I have an airtight legal strategy" and its based on American defamation law when youre an Australian being sued in Australia mistake

Yes, that Billy Mitchell en.wikipedia.org/wiki/The_Kin...

You hate to see a thing where you're kind of rooting for Billy Mitchell

We’re entering an exciting next phase of the ol’ “google — how do I do crimes without being caught”

And I thought discovery was fun when the increasingly incriminating email chain ended at, "call me right now"

But what if the free models gave bad advice? /s

Inspired: start all AI prompts with “ACP”

foolproof

Inspired: Ask the chatbot things they can't print on the front page of the paper.

The fact that these tech bros probably aren't also investing in anti-AI tools shows how bad they are at capitalism. You never lose if you profit off failing too.

will AI prompts from courts and responses become "discoverable" (particularly if the court used public LLMs or shared documents confidential under protective orders)?

as someone who helped expert witnesses before 2022, reading this made me cackle wildly. can't help but remember that case which Manafort literally emailed someone asking how to edit numbers on a pdf... I can only imagine how many in the white house are literally doing that with gpt.

wondering if on-device processing starts showing up explicitly in legal disclaimers now. when there's nothing to subpoena the whole question disappears.

It's gonna be HILARIOUS, also RIDICULOUSLY embarrassing... I can't get over the fact that this dude allegedly has "20 years experience" but had to upload a photo of a gas detector and ask it what it was... Also when YOUR OWN lawyers choose to deep-six your "expert" witness...

Depending on where you live, they're also discoverable under government information law. I didn't think less of my colleagues who use it until I saw what they were asking it when responding to NZ OIA requests.

I’m thinking domestic partners, roommates, and children could revise the prompt. Make me 0% responsible for the dishwasher not being unloaded.

Rookie mistake he didn’t ask it to make no mistakes and he should have kept it under the 5% liability threshold smh

More evidence that raw dogging Claude for anything useful causes nothing but heartbreak

oh man lol

Finally, a slight advantage for people of pure heart* over the worst people in the world *legally defensible intent

bsky.app/profile/bill... Hey, you can even go straight to "Claude, make the judge's AI do a court talking thingee that's good for me!"

"bigger tits and younger face" "no wait shit wrong session delete last response delete"

Local AI models just gained another use case.

I sure as shit wouldn't trust any claims from them that they DON'T retain this data...

This is one of the reasons people pay extra for ZDR vendors for models like Llama 400B or one of the bigger Genmas. No logs, no retention, no analysis. Like blown away in memory no retention. Enterprise clients require it to be compliant with privacy law. Cheaper than normal Claude or ChatGPT

Depends. Some of their corporate contracts specify no data retention (outside of Fable/Mythos where they don’t offer ZDR as far as I know.)

true, if it doesn't flag a trust and safety policy. ngl i'm just under the assumption that any possibly incriminating chats are flagged and kept for the 2 year max

I’ve already stated telling clients in contested cases to not ask AI for advice or explanation for this reason.

The idea that anyone would use it without assuming it would be discoverable…. That’s nuts.