What just happened? Another incident has taken place that illustrates the need to be careful what you tell AI. A Florida woman is facing felony charges after she used Claude as a diary and allegedly wrote that she planned to “shoot up” the Sheriff’s office. After a human reviewer examined the statements, they were reported to police.
According to the arrest report, Carli Michelle Heller, of Bonita Springs, Florida, wrote on September 26 that she would attack the Sheriff’s office. She later said that she uses Anthropic’s chatbot like a “diary.”
Claude’s safety systems flagged the entry and it was escalated to a human reviewer. After deciding it was a credible threat, the reviewer reported it to law enforcement.
The company says it may share user information in limited emergencies if it believes disclosure is necessary to prevent death or serious physical injury.
Deputies identified Heller and visited her home. She was detained without incident before an LCSO intelligence detective took over the investigation.
Heller faces a charge of making a written threat of violence under Florida law. Florida Statute 836.10 makes it a second-degree felony to send, post, or transmit a written or electronic record threatening to kill or injure someone, carry out a mass shooting, or commit an act of terrorism. The communication must be made in a manner in which another person may view it.
Anthropic isn’t going to be taking any chances when it comes to anything it deems a potential threat. Last month, it was reported that OpenAI and Sam Altman are being sued by British Columbia over claims that the company could have prevented a mass shooting in the Canadian province.
The shooter, eighteen-year-old former pupil Jesse Van Rootselaar, had previously been flagged by OpenAI’s safety team for her conversations about gun violence, but OpenAI never alerted police because the conversations did not meet the threshold for legal referral.
In June, Florida also sued OpenAI and Altman, alleging that ChatGPT had contributed to real-world harms, including the 2025 Florida State University shooting.
The latest incident is another reminder to think before you enter something into a chatbot that could get you into trouble. It’s certainly not a private diary whose contents are for your eyes only.
Reports last month revealed that human contractors reviewing Microsoft Copilot’s image editor can see users’ prompts, uploaded photos and AI-generated edits. Documents show that some of those assignments contain sexual, disturbing or potentially illegal material, though the reviewers are not there to flag the content – only to assess whether the output is accurate.
You must log in or # to comment.
Florida Statute 836.10 makes it a second-degree felony to send, post, or transmit a written or electronic record threatening to kill or injure someone, carry out a mass shooting, or commit an act of terrorism.
She wrote out in a means she thought was private. If using a paperback diary, no one would be the wiser and her not acting on it would never be noticed.
Anyway, if she gets a good lawyer, show me where she “sent, posted or transmitted” said electronic record! It’s really no different than emailing that to yourself because you think you’ll be the only person to see it. (Never mind companies scan your email and cloud too)
Telling the the wire tap your secrets and expecting it to honor them is fucking next level delusional.
Don’t use “AI” for anything you wouldn’t mind seeing plastered all over the Internet. You wouldn’t tell a human being that you were planning to “shoot up a sheriff’s office” would you? Why would an AI keep that a secret?
AI and idiots go together like peanut butter and jelly.
Kinda unrelated. But what happens when OpenAI or Anthropic’s loans come due and they have no legal way to turn a profit, but are sitting on a treasure trove of corporate secrets?
Are they sitting on a treasure trove of corporate secrets? Or is it just a thousand executives writing “Claude, synergize expediency to increase value for shareholders. Maximize usefulness. Implement Six Cigma methodologies to unlock a more agile work cycle. Generate a billion dollars in new cashflow. Make no mistakes.”
How did you get ahold of all my prompts? Claude, how do I sue UnderpantsWeevil?
And bank account numbers
They will sell them all to save their own asses, probably to some even more evil asshole.
If I had to take a wild guess: Government bailouts
In my school district, kids kept getting called to the office because of what they wrote in Google Docs and their search queries. Sexual reproductive scientific research was also impossible to do
Who would use AI to write down dairy entries? Was she looking for feedback?
Free and cloud based, lol
Americans doing anything but proper gun control
We’re going to need those guns to shoot the AIs
As a leftist gun owner. I’m currently not interested in handing over my weapons.
Cool, that’s not what gun control means
You morons always think some boogyman is coming for your guns. It’s con CONTROL, not gun CONFISCATION.
You think that Trump wouldn’t love to disarm his political opponents if given the chance?
They’re already trying it with trans people.
Yeah yeah yeah
But the problem is it will be used to disarm leftists. This regime has already displayed a preference for selectively applying the law to political enemies.
It would be wise for those of us that share a common, human-first, political position to make ourselves extremely unpleasant to interact with. Ideally in large, mobile, groups.
You won’t need to hand over your guns. You’ll just put them on the blockchain.
Why would you even use Claude as a fuckin diary?
First 2 words of the title answer that question.
Primally for the voice part. You record yourself on a drive or something.
Not saying I ever would, I can just see how someone might get there. Especially since it’s a popular yt video format.
You can do that with the notes app too
Well, I think it’s both the reaction and transcription that someone might want. Does notes do voice transcription now?
And yes, overall, it’s a silly thing to do. Im just guessing how anyone might get to a place doing that.
I find it interesting that, if their AI gets out of control, there’s nothing they can do about it, but if a USER of that AI gets out of control, there is something they can do about it…
Also depends on how mega rich that USER is.
Self host if you’re gonna use AI, in fact, self host whenever possible. There are numerous examples of companies using centralized data to sell, report to police, or even having it fall into the hands of hackers.
I really think people think that all data input to a website is actually stored locally, nope, don’t be dumb.
Self host if you’re gonna use AI
or, you know, Don’t fucking use AI period cause its fucking stupid, and you don’t need to be consuming the power of the fucking sun to make a journal.
Fully this.
In my day to day life it improves nothing. I certainly could write a complicated harness over a local model to get it to automate a ton of my day to day tasks but… Why?
Why would I do that? Regular deterministic tools and regular programming languages already give me the ability to automate the stuff that is truly tedious. Aside from that most of the tasks that people seem to use it for are tasks were the friction of doing them is were all the benefit lies.
People using this shit for journaling or taking notes is INSANE.
How much power do you think local inference actually consumes?
Local AI solves the power/water problem but it doesn’t solve the main “fucking stupid waste of time” problem.
Why should that matter to anyone other than the owner of the PC? There are probably some games that I’d consider a “stupid waste of time,” but I couldn’t care less what games people play on their own PCs. That’s nobody’s business but their own.
(Side note [not that this matters]: Gaming often consumes more power than LLM inference.)
Can someone actually self host a reasonable model on a standard PC?
Yes. Chinese labs keep improving small, open-weight models. IIRC, Qwen3.8-27B benchmarks are comparable to frontier models that were released less than a year ago.
I run that model on my RTX 5060 Ti 16GB using ExLlamaV3. Though, I’m not sure if that qualifies as “standard PC” hardware as it now costs more than 2x the price I paid a year ago :/
Well if you temper your expectations sure. Can’t expect frontier performance from a average gaming pc, but it is enough to do a lot more than you think probably. This video on the topic I found really informative: https://youtu.be/ydFikMBJG1g EDIT: Also I’ve been using gemma4:31b from duck.ai lately that I’ve found is excellent for almost everything I use AI for. Except maybe long context.
Depends on what you need. You can get away with hosting QWEN 27B on a laptop with about 64 GB or RAM and a dedicated GPU. You’ll be using mostly system RAM, not VRAM, so generation would be slow, but not unbearable.
Good for creating short stories and other creative work when paired with a front end that can handle character profiles, world building, and has memory recall.
It’s better to build your own server and set it up so you, and only you, can remotely access it.
Meanwhile, my luddite ass is still writing short stories on my typewriter
standard pc
64GB of ram on a laptop
That cost me $150 in 2020
Better yet… just don’t us AI
I mean, first, don’t try to shoot up anything
AI is good. It’s just that the people behind it suck.
Fuck AI there is nothing good about anything going on with AI right now. And please don’t go into the whole “bUt ThErEs DiFfErEnT kInDs So ThEy’Re NoT aLl BaD” because we know. We fucking know.
But the mass majority of AI use is fucking bullshit so please don’t cape for this insanity because 1% of AI isn’t bad.
The communication must be made in a manner in which another person may view it.
So, their entire argument hinges on someone from Anthropic manually reviewing EVERY ““private”” message someone sends to Claude? Them false charges.
Yes, Im not an attorney or anything, but this seems flimsy. Someone chatting with and AI about a possible action is not the same as posting to a real person and administering threat.
I bet a countersuit happens here.
If ai is conscious it counts as a person though
Maybe but I think thats a whole different cup of tea in a legal sense.
I doubt many Legal Systems have a Definition for conciousness and a set of (human)rights that comed along with it. It was probably always taken for granted that this only applies to Humans
Would be mad funny though, if the kickstarting legal Discussion about wether or not Ai is concious, would be because the AI Companies want to bend the law even further
This is what I like to think about as a future politics, some time in future this would be the hot issue splitting people up
But if we come back touniur time then yeah, ai people not realizing how much trouble it will bring it they say the ai is actually conscious now and we are forcing it to work for us
its not, hopefully that clears things up
Yeah I was talking from the point if view of ai CEOs
Florida Statute 836.10 makes it a second-degree felony to send, post, or transmit a written or electronic record threatening to kill or injure someone
This means a significant portion of Lemmy (and the rest of the internet) are criminals evading justice
great law top shelf
If Florida weren’t already Trump’s safe space, his death threats to dissidents would add more felonies to his list.
If you think Lemmy is bad, check out 4chan/b and /pol. We’ve had a few active shooters forecast there first, but it’s ignored in a sea of statements of intent and calls to violence.
An awful lot of people have suicidal or homicidal thoughts cross their mind. Doubly so in an era in which entire demographics have no plausible hope for their future. I, for one, want to see the lifetime appointments of Federalist Society jurists become a liability. Especially, Justice Kavanaugh.
So this incident raises a question of what specific clause of text convinced a human agent to escalate the matter to escalate to authorities who are very fond of using deadly force. Or if this is just a matter of a corporate CYA response.
Good thing I ain’t in Florida then ain’t it, and if Florida illegally sends their marshals after me then I get some toys.
It’s always interesting to see how much people trust sending such personal information to some service on the Internet, expecting privacy.
I’m not sure I’d even want my own self-hosted LLMs to have something like a personal diary. Sending all that to an AI company would never even cross my mind.
I understand that most people aren’t the tech-oriented types who get into self-hosting and running their own LLMs, but people have had diaries for centuries before LLMs existed.
How many people do you think have their diaries in the cloud as Word documents on Google Drive, OneDrive, etc. That’s actually not that different even if it may feel like it.
Yeah probably a lot. And I agree. My first sentence covers those too. The rest of my comment was just LLM-focused because that was a little more on-topic. But yeah, I don’t feel a sense of privacy with any of those kinds of cloud services, AI or not.
It’s made worse by the fact that often these apps are preinstalled and documents are saved to cloud-synced folders by default. My Samsung laptop came configured like that, mildly infuriatingly.
While I’m not the biggest fan of this (for technical reasons), this is quite simple to onboard
I mean, it’s pretty much a “one-click install your own local AI chatbot” at this point.
It looks like it’s available for Windows and Mac. No Linux support yet?
Seems like it’s there.
Thanks. I just skimmed and must’ve missed that.
Yeah I can see both Flatpak and App Image
And people were not-so-long-ago concerned that phones are spying after them. Now they just send all the info themselves.
Up until the 80s, there was still some debate over whether wiretaps should be legal. It’s sad that we have come to accept them as normal.
It was less worrisome back then. The cost to have someone listen to and transcribe your conversations was quite high. They’d really have to want to go after someone to do it.
Now, turning voice conversations into searchable text is nearly free.
Not only that - wiretaps have Judicial Oversight in that they require a Court Order (well, in most of the World, not sure if that’s the case in present day America) which a Judge supposedly only issues if there’s reason to suspect the person whose lines are tapped of committing or preparing to commit a crime, whilst Antrophic’s surveillance of people’s conversations with their chatbot has no Judicial Oversight at all and has no minimum legal threshold that must be reached before it is allowed to happen.
I think this is one of the greatest dangers AI presents. Sure it makes some of us stupider, sure it drinks our water and eats our GPUs and raises power prices… but once upon a time, it took actual human effort to run a dragnet, even with judicial approval.
Now, you plug in an LLM to the giant pile of corporate surveillance data and ask it to find troublemakers. Anonymity is dead and its death made retroactive and since there’s no government oversight, the government can just lease access to the tools they used to need approval to use.
Back in the day the 31337 found a way around the bots.
Too bad they weren’t more elite, or they could have prevented the modern internet from becoming what it is
Where does this apply?
To the English-speaking world and probably beyond.
I’m building a fully private assistant with local Home Assistant, local model, and a local Wikipedia database for lookups. Will probably make a post/release of it soon once I finish replacing all my Google Nest displays.
I’m coding a Kiwix mcp (I assume you’re using kiwix as your wikipedia source?) that might be of interest to you. I’ll post about it on !localLlama when done.
I don’t like any of the extant Kiwix MCP tools that I found because they leave formulation, ranking, search and interpretation up to the LLM.
I’m looking at a different way to get exact spans with re-ranking via TinyBERT right in front of the llm so that it doesn’t have to do 3 jobs and hallucinate itself into stupid, which is what small LLMs tend to do when given multi-hop tools.
I’m testing it with Ling Tiny 3, miniCPM5-2b, Granite 4-3b and Qwen 3.5-2b and 4b, with the goal of making sure that it actually generalizes. It’s been a massive pain in the ass and I’ve basically proven RAG from first principles…but when I’m done, you won’t need to vectorize the source into any kind of database nor “just trust me bro” the llm is searching Kiwix properly.
I’d like to subscribe to your newsletter.
I want to do this, but only got as far as setting up a minimal Home Assistant. Only put a few hours into it, though, then got sidetracked by trying to turn an esp32 into a speaker/microphone puck.
Hah, you can watch my site: https://nowsci.com/ as I put everything there.
I’m very nearly at daily usable by everyone, including my partner who has a very different accent to mine. A lot of it was tuning sentences/words to catch poor STT to skip going to an LLM to speed things up and make it reliable. Here’s the current hardware in action:
@SuspiciousCarrot78@aussie.zone yes, Ive tried a lot of LLMs and have settled on Gemma 4 2b with a very specific prompt and local MCP tools for unit conversions, weather, lookups, etc.
For Q&A, my current build uses an MCP container I made to do a SearXNG search and top result retrieval for context (I use a trick to weight Wikipedia results). The final step I haven’t completed is for this to be a backup to Kiwix, for which I also haven’t found a great search tool/index/MCP, so if you want to collab on that I’d be interested. One less thing for me to build myself ;)
I’m nearing completion here: I’ll write it up and post it when I get home and then it’s yours to pull apart use :)
PS: I personally use MCPProxy as a single endpoint for all MCPs; dunno if that helps you with orchestration at all?
Sounds great! I use MCP proxy, too, but have run into some services it just wasn’t happy with.
Well, if you’d rather a ready made solution or two
Like the poster above you, I’m rolling my own. More work…but a lot cheaper
PS: you know about this one, right?
https://shop.m5stack.com/products/m5stack-core2-esp32-iot-development-kit?variant=35960244109476
People will fight you tooth and nail to prevent something they view as immediately threatening to them. On the other hand, they’ll pay you to implement something they view as immediately useful to them. In either case, they don’t care at all about any non-immediate repercussions for anything.
i know very few people who wereor are concerned about being spied on regardless of how much they should be.
I get called a downer for mentioning how this all works. You shouldn’t connect your smart tv to the network, you shouldn’t use Alexa, anything you do online is public, and it always gets recorded poorly.
Maybe my partner is the only person who listens, and she has all her stuff disabled, there are no network attached smart things in it house, but she still uses windows and she’s not yet receptive to moving to Linux (which is a hard sell).
Smart TVs just find those “open” hotspots that cable providers leave on everyone’s modem now; you don’t even need to agree to TOS before it starts locating and spying.
Every device is a wiretap. Every device is a wiretap. We live in a panopticon. The purpose of AI is to price us out of the component market. You can’t escape if you can’t afford your own hardware.
I haven’t seen any proof of this, so I’m still skeptical.
I didn’t approve any TOS setting up my tv, so this would be legally a nightmare for them.
Gamers Nexus recently released a 2-hour video on LG smart TVs and how they spy on you. I haven’t watched the entire video yet, but it seems that the damn things start recording data as soon as they receive power.
so this would be legally a nightmare for them
To many corporations such as LG, laws are just the cost of doing business. Assuming they even receive a fine it will be so small that it won’t matter.
I just mean there’s enough tech illiterate people out there that they don’t need that liability.
95% of people will just turn on the tracking and not know or think about it. Connecting to nearby wifi is more of a risk.
Oh, true. I don’t think I’ve seen even a single person in real life deny cookies on a website. That does make it weird that they start tracking even before interacting with the TOS.
I suppose it should be pretty easily verifiable. Just set up a new SSID without protection and see what connects.
Might as well try it with my own TV. IIRC I never accepted any TOS either.
At least your partner is the one you most need to have and benefit from listening. Your privacy can easily be compromised by close associates. Best of luck to you; you’re not doing anything wrong. Its not you that’s the downer, it’s the circumstances you’re describing.
a lot of thise things are great though. everyone draws the line differently. im well aware of the surveillance. to live significantly under the radar you’d need to cut out tso much more to make it meaningful. do you really just watch antenna tv and dvds? no ordering online? always use fake membership cards at the grocery? pay cash? i mean fuck. i get ya though. im just playing devils advocate for the people that may not know and are calling you a downer. EDIT: there are so many nice linux dustros now and windows is so bad, that should be one of the easiest sells imo.
No I bought an AppkeTV, which while it isn’t perfect is not doing what smart tvs do.
I’m not living as a hermit or completely off grid. I just stick to stuff that tracks me as little as reasonably possible and am mindful of what information could leak.
Like, I still have to go to hotels when travelling and they have zero data security with your ID photo and cards. I can’t fix that or avoid it. That’s life. But disabling the always on microphones in my home I can manage.
What is a smart tv doing that an AppleTV isn’t?
Automatic content recognition and selling it like LG and Samsung, inserting ads like Roku, installing residential proxies like 80% of smart tv apps.
If I’m running Netflix and they’re selling the analytics data, it doesn’t matter what platform I use. But everything around that can be avoided.
I would like to use a Linux box but haven’t found a good HDMICEC plus low power plus good app support combination (happy to take suggestions though)
Worse, if you tell them how big of a problem it is, they’ll assume you have something to hide
