Imagine this scenario: you’re worried you may have committed a crime, so you turn to a trusted advisor — OpenAI’s blockbuster ChatGPT, say — to describe what you did and get its advice.
This isn’t remotely far-fetched; lots of people are already getting legal assistance from AI, on everything from divorce proceedings to parking violations. Because people are amazingly stupid, it’s almost certain that people have already asked the bot for advice about enormously consequential questions about, say, murder or drug charges.
According to OpenAI CEO Sam Altman, anyone’s who’s done so has made a massive error — because unlike a human lawyer with whom you enjoy sweeping confidentiality protections, ChatGPT conversations can be used against you in court.
Now, if your lawyer (because they fired their paralegals) asks ChatGPT something while working on your case, can their queries be used against you?
Probably not, but the false case citations that it gives you back can get you in trouble with the court if you don’t double check them. This keeps happening, even though every lawyer in the country should have caught wind of the first guy who got sanctioned for it. Seriously, are there no trade publications for the legal profession? I read up on all kinds of gotchas to avoid in my industry (tech).
Very good point. It seems crazy that this keeps happening, right? My current favourite hypothesis is from “Power and Progress” by Daron Acemoglu and Simon Johnson:
If everybody becomes convinced that artificial-intelligence technologies are needed, then businesses will invest in artificial intelligence, even when there are alternative ways of organizing production that could be more beneficial.
Add to that sunk cost (these firms invested in this tech, and maybe fired the paralegals that used to do this work, so they need to use the tech), and fundamental attribution error (those other lawyers failed using AI because of something fundamentally a part of their selves, I am only fail when there are external factors getting in the way of my self) and you get a recipe for seemingly irrational behaviour on repeat.
You mean I shouldn’t sign up for an account with somebody else’s chatbot using credentials traceable to me and confess crimes to it?
This is why I only confess my crimes to my local LLM.
What’s the name and IP of your LLM? I also want to confess some crimes.
It’s at http://127.0.0.1:5001/. It’s my sex box, though. And her name is Sony.
FBI: No, it’s fine, they have a privacy policy.
Lol. LMAO, even.
Using a search engine for an answer now could get you in that same trap then no? Even DDG has an AI answer generated automatically.
We didn’t need AI for Google search queries to show up in murder cases where people searched for how to clean blood / dna with bleach, how to dispose of bodies, and other really stupid questions. From their home computers. Pretty much right around the time that someone close to them went missing. It’s all timestamped.
I would imagine that the same is true of search engine queries, and the collection of search engine queries is gonna be a lot larger than ChatGPT prompts.
I’ve seen articles mention search history as evidence for years, so no need to imagine.
I would mention that I use Kagi as a search engine, which doesn’t retain search queries (outside a subset for a limited period of time, like a percentage retained for 7 days to counter DDoS attacks). They also have functionality to let people — people who are more paranoid than I am — pay anonymously via cryptocurrency and use search tokens that don’t link individual searches to each other.
Maximizing anonymity with Kagi
We strive to give our customers the possibility to maximize their anonymity. Users who want provable anonymity guarantees may access our service by:
- Creating an account with a pseudonymous email address
- Paying for their plan using cryptocurrency
- Accessing our services via Tor service
- Anonymously authenticating using the Privacy Pass protocol
https://blog.kagi.com/kagi-privacy-pass
In general terms, Privacy Pass allows “Clients” (generally users) to authenticate to “Servers” (like Kagi) in such a way that while the Server can verify that the connecting Client has the right to access its services, it cannot determine which of its rightful Clients is actually connecting.
I’d also add that, search query and AI prompts aside, one’s aggregate YouTube history is also likely to have privacy implications, as I expect that most people have watched a lot of content on YouTube and done many searches on it. In 2025, I can’t suggest a reasonable, privacy-oriented drop-in alternative for that, though.
EDIT: Social media is its own can of privacy worms, but at least there people basically understand that they’re putting content out there for the world to see, albeit maybe wanting to do so pseudonymously.
EDIT2: Actually, I haven’t been using it, but Kagi does have a video search, and a bit of experimentation shows that it does appear to index YouTube. I guess I could use that to hide search queries, though obviously YouTube will still have a “videos watched” history, as one would still connect to YouTube for a video itself. And it’s gonna come with some limitations; NewPipe and similar mobile clients don’t have functionality to issue search queries to anything but YouTube directly, so one wouldn’t be able to use a mobile client for searches. I also don’t know whether they permit filtering on everything that YouTube does (or, if they index multiple video sources, whether it’s even possible to filter things on all those criteria; different video services may not expose the same information).
EDIT3: It also appears to only return 48 results per search, unlike YouTube’s search Web UI, where I believe that you can just keep paging through more results as long as you want.
EDIT4: Ah, they show what they index in the search options, since they let you choose which source the videos are from. Apparently it’s YouTube, Vimeo, TED, TikTok, Twitch, Daily Motion, and PeerTube. Huh. I didn’t even know that it was possible to search Vimeo at all. Last time I went looking for a YouTube alternative, I remember looking at it, seeing that the main page had no search form or list of videos, and thinking that there wasn’t any way to search it at all.
Startpage doesn’t track, and you can view result pages via their anonymous proxy. Use that with tor or a VPN and you’re set. YouTube doesn’t like startpages anon proxy though.
Duckduckgo doesn’t track either, and you can view youtube videos in ddg’s site.
https://www.startpage.com/en/privacy-policy
https://duckduckgo.com/privacy
deleted by creator
99.9% of today’s internet is encrypted by default - https.
DDG definitely does. (Though you want to change the settings on the site from GET to POST for privacy)
The current tls standards in Https/3 /quic is especially resilient for standard encrypted web connections.
So, what you do on a site can’t be seen by anyone other than you and the site.
What a site collects is what you allow them to have. And trusting (or not) to believe their policies.
If you use encrypted DNS (doh, dot, dnscrypt, etc.) your ISP can’t see what you’re looking up. If you use encrypted AND oblivious doh, or anonymized dnscrypt, nobody can tell what sites your DNS is looking for and the DNS resolver itself can’t tell where the request you make comes from.
Ech/esni is the newest standard for DNS where nobody can even tell what sites you are making a secure connection to. (The IP can be seen, but not the domain).
With that, using a half decent secops and basic isolation of accounts, profiles, use cases, and housekeeping, and a VPN/proxy/tunnel, your internet use / history isn’t difficult to manage to keep private.
If I’m coming from a different IP, and a different browser (or the exact same as a million others - tor, mullvad) on a different profile, in a container/isolation/vm, with a different fingerprint, via a different DNS, spoofed device identifiers, and privacy scripts/blockers, etc, the likelihood of my use being tracked and used is exceptionally limited.
A browser for porn A browser for local A browser for social A browser for porn
Wow. People really have become stupid. Asking an AI for help in medical and legal questions is among the dumbest thing one can do.
If you committed a crime and confessed. gPT didn’t fuck you. You fucked yourself up. Asking Google would have been just as bad.
Accidentally doomed yourself, as opposed to intentionally doing it by getting legal advice from ChatGPT in the first place.
This has been the case for a long time. Plenty of folks have had search history used as evidence.
Sensationalism at its best.
What do you know? You’re just an AI
Yeah, this was famously a huge part of the Casey Anthony trial.
Dear ChatGPT, I want to build a spaceship fleet and conquer the Andromeda galaxy and use it to power my AI data center. Can I get in trouble for that? Thanks. Sincerely, LLMAO.
“Dear ChatGPT. I killed a hooker. What are the best ways to dispose of the body? And while we’re at it, can you please give me tips on how to avoid getting caught trafficking meth? Thanks so much!”
-People who deserve to get caught.
…In Minecraft.
Ew, Theo Von
Rapp Snitch Knishes by MF DOOM comes to mind
Rap snitches, telling all their business Sit in the court and be their own star witness “Do you see the perpetrator?” “Yeah, I’m right here” Fuck around, get the whole label sent up for years
Love that song!
According to OpenAI CEO Sam Altman
Why would I care about Altman’s opinion on this?
Because he’s the one handing over conversation records to the police.
In this case, he’s right.
So? He’s not a lawyer. If this is obvious to non-lawyers just say it. Or do they have to report everything he says on any topic?
“According to OpenAI CEO Sam Altman pizza is good!”.
If this is obvious to non-lawyers just say it.
That’s what he did.
