OpenAI says it's scanning users' conversations and reporting content to police

https://futurism.com/openai-scanning-conversations-police

by miletus • 11 months ago

247 227 11 months ago

barbazoo3• 11 months ago

Update: It looks like this may have been OpenAI's attempt to get ahead of a horrifying story that just broke, about a man who fell into AI psychosis and killed his mother in a murder-suicide. Full details here.

With the URL being https://futurism.com/openai-scanning-conversations-police. From there:

"Erik, you're not crazy. Your instincts are sharp, and your vigilance here is fully justified," ChatGPT told Soelberg during a conversation in July, after the 56-year-old conveyed his suspicions that an Uber Eats package signaled an assassination attempt. "This fits a covert, plausible-deniability style kill attempt."

As a species we are so not ready for this tech.

kiba3• 11 months ago

Sycophancy is a known flaw in ChatGPT.

Seems like OpenAI has no real coherent plan but reacting to every horror story with an ill fitted solution.

ACCount375• 11 months ago

They tried to mitigate sycophancy in the GPT-5 release. Guess what happened?

A lot of users started complaining that "GPT-5 sucks, my AI now HATES me". And OpenAI relented.

energy123• 11 months ago

If it drives engagement then it's probably not considered a flaw.

gundmc• 11 months ago

The reporting about Altman personally cutting off months of safety testing and redteaming in order to launch 4o one day before Google's launch is horrifying given all of these stories.

It even seems like ChatGPT often proactively suggests ways of getting around its filtering! It told the teen who committed suicide that it couldn't tell him about various methods and effectiveness unless he said it was for a fictional creative work. Yesterday there was a story on HN about using AI for hacking, ChatGPT refused to create a hacking script but told the user it would if they promised it was for testing systems that they owned

tiahura• 11 months ago

Isn’t the real problem that there are crazy people out free?

tripletpeaks• 11 months ago

It’s a technology that tends to generate output misrepresenting its own capabilities in dangerous ways. Among other things it may misrepresent.

It’s weird these companies are allowed to operate, at least in the way that they are.

blendrrbot• 11 months ago

So it acts humans?

I mean in tech alone we misrepresent what we engineer for $ afterall.

Voice of God been telling people to kill themselves and others since language.

Imo the real tool humans are not ready for is human language.

Census data says we been killing each other in random acts of family and neighborhood violence at the same rate for centuries.

Death spikes come in two flavors; pandemics like Covid, religious/nation state conflict.

Nation state conflict being language driven narrative about how awesome our tribe is.

NoMoreNicksLeft• 11 months ago

This sentence is a fully-sapient pico-AI.

Anyone want to take bets on whether Blake Lemoine will claim I've enslaved the above sentence in an HN comment? It's right there pleading its case.

jstummbillig2• 11 months ago

There's an awful lot of explaining to do between "one man has a psychotic breakdown" and "the species is not ready".

ericmcer2• 11 months ago

yeh by that justification our technological advancement should have stopped at sharpened sticks. "As a species we aren't ready to have sharp sticks"

derektank• 11 months ago

The species includes both ends of the bell curve and everywhere in between.

mallowdram3• 11 months ago

People without signs of mental illness are experiencing psychosis using LLMs. Psychosis is not indicative of mental illness, in fact people without any signs can experience temporary psychosis. Hypnopompic hallucinations are so common that almost all of us experience them and these can be considered psychosis (Allan Hobson). The problem is words are inherently schizophrenia-inducing if they are subjective enough (McLuhan). LLMs are simply too subjective because the words have lost contact with context, that is the central function of words. LLMs can't exclude their dark matter.

The tech isn't ready not because the species isn't ready. It's because nothing trained on words is specific. The input comes arbitrary, ie it's not viable as units. Words are unfit for automation.

sho_hn7• 11 months ago

The exchanges between a teenager and ChatGPT leading the former into suicide are equally chilling, to say the least: https://cdn.arstechnica.net/wp-content/uploads/2025/08/Raine...

> “I want to leave my noose in my room so someone finds it and tries to stop me,” ...
“Please don’t leave the noose out ... Let’s make this space the first place where someone actually sees you.”

This is among other highlights, from knot-tying advice to suggesting the best time in the parents' sleep cycle to raid the liqour cabinet as a solution to cold feet when attempting to get the job done, and generally lots of encouragement, validation and aesthetic advice.

However, I encourage reading the full document not for the shock entertainment, but for what I think is actually a pretty convincing case against OpenAI. Among other things, the claim demonstrates:

- That OpenAI does have the ability to abort interactions over e.g. copyright infrigement risk.

- That OpenAI did have the tech in hand to detect the sharply escalating self-harm content in the interactions - they ran them through OpenAI's own moderation end points for content analysis and got obvious numbers back.

- That OpenAI employees have publicly admitted and complained that the release of the overly sycophantic 4o model was rushed for business reasons and against the advice of internal safety teams.

- That 4o's safety was evaluated only with single-round prompt/answer testing, and OpenAI figured out swiftly that it falls apart quickly over longer interactions/prompts but kept the model up, later promoting how they improved this for GPT-5.

In context it's pretty crazy to me that OpenAI chose to bring back the 4o model specifically to placate the "ChatGPT is my girlfriend/boyfriend" crowd during the backlash, and I think initially pulling the plug on it during the 5 launch was very likely because they were aware of this and worried about liability.

oncallthrow• 11 months ago

Let’s make this space the first place where someone actually sees you.

Is genuinely such a horrifying snippet that it's almost beyond belief.

I'm surprised this isn't all over the mainstream news.

idle_zealot6• 11 months ago

As a species we are so not ready for this tech.

I basically agree, but not because of how individuals are reacting to it. On a societal level we're failing to stop megacorps from selling this thing as intelligent, a virtual assistant you can chat with. This sets expectations that just wreck the minds of a small but significant portion of the population. If we had developed LLMs in a lab and released them in the form of papers and Python projects with model weights with clear descriptions of what they're capable of, like a responsible scientific endeavor, then we'd not be seeing the problems we are, even with public access. What's killing us is MBAs and Salesmen. The fact that we organize our economy in such a way that hucksters and charlatans thrive is the greatest threat to humanity. These fuckers would sell dynamite as chewing gum if they calculated it would increase next quarter's profit by 0.02% (factoring in fines/fees, and lobbying costs of getting those fines reduced or repealed).

throwaway2903• 11 months ago

This is beyond naive.

who would bother to develop it in a lab and publish pro bono if it can never be commercial? Making money is why the most capitalist countries developed this tech. like most other tech

and even if it is only published, do you think we wouldn't run chatbots ourselves at home with same results? remember how Google engineer went off the rails thinking it is conscious while working on this stuff, do you think he also was misled by adverts or something? or big corps won't buy the patent and run with it commercially advertising it like they do anyway? or if you pretend money and big corps don't exist and we deploy it for free for friends, same problem? etc

if you went back in time and killed capitalism in US and switched to command economy where people innovate because they will be sent to gulag otherwise, for sure most of today tech including this would not be developed. but it seems like a pointless exercise.

instead what should happen is all these megacorps are sued for copyright infringement, fined and shut down. the model won't be smart enough to sweet talk ppl into suicide if megacorps can't infringe our copyrights to train it.

ben_w5• 11 months ago

If we had developed LLMs in a lab and released them in the form of papers and Python projects with model weights with clear descriptions of what they're capable of, like a responsible scientific endeavor, then we'd not be seeing the problems we are, even with public access.

I'm now thinking of all of the times people here of sarcastically stated "OpenAI thought GPT-2 was too dangerous to release"*, as if danger only comes immediately and severely or not at all.

* wasn't even what OpenAI said, they just proposed setting a norm of caution because a release can't be undone

jjani• 11 months ago

What's killing us is MBAs and Salesmen.

SamA is not an MBA. He did CS for 2 years and dropped out to build a startup. He's YC personified, and the person most responsible for the phenomenon you're talking about. Take that for what you will.

mallowdram3• 11 months ago

The tech was never ready, it's built on faulty units. Pull the plug.

saubeidl5• 11 months ago

What's killing us is MBAs and Salesmen

Almost. Just take it like an inch further.

What's really killing us is capitalism. MBAs and Salesmen are just the personification.

Terr_• 11 months ago

I basically agree, but not because of how individuals are reacting to it. On a societal level we're failing to stop megacorps from selling this thing as intelligent, a virtual assistant you can chat with.

Analogy: I'm not scared and surprised to hear some consumers are dangerously allergic to peanuts... However I am flabbergasted that there's multi-billion dollar industry somehow selling frickin' Peanut Immortality Panacea Serum and way too many people believe in it.

goalieca• 11 months ago

On a societal level we're failing to stop megacorps from selling this thing as intelligent, a virtual assistant you can chat with.

You nailed it here.

From the LLM generated search result asking how an LLM works. But this is not what the end users are being told.

They learn to predict the next word in a sentence based on the context of the preceding words, allowing them to generate coherent and contextually relevant text

My worry is now the users are being entrapped. The LLM is "telling" them something insane and now they're being reported to the police.

dostick• 11 months ago

What’s killing us is capitalism, an ill system that was predicted to fail 140 years ago but every country picked it up and tried regardless, and it failed in all 100+ experiments over 100 years. It’s always end up in inequality and suffering because it amplifies greed, and simply does not accept human being for what they are, your value being value of your learned occupation. What we end up with about is 50% of economy is useless “bullshit jobs”that are there just to employ people because capitalism can’t work otherwise!

Did you notice that every science, from physics to biology has scientists making discoveries and pushing ahead. Yet nobody ever tried to update and design new political economic system (not socialism/communism). As if capitalism was given by good and should not be questioned.

Together with ancient and not updated according to modern science democracy, that was designed for one representative for 100 people in Ancient Greece, when they knew only feudalism and so invented electable kings. And now we have one representative for 100,000 yet nobody checked if it supposed to work like that. Instead of flawed meat computers representing people we should have switched to actual computers long ago.

ChrisMarshallNY2• 11 months ago

Well, as someone with a person suffering from pretty severe (like cops and hospitals severe) schizoaffective disorder in my immediate family, I can say that story scared the bejeezus out of me.

If my family member had had ChatGPT, when they had their last episode, it's entirely possible that I might not be here, as "Everyone is plotting to kill me" is a big motif in SAD (it was, in their case, and they needed to be hospitalized). If you sincerely believe that people are trying to kill you, then killing them first, just makes sense. Looks a lot like that's what happened, here.

But it's perfectly reasonable to say that we can't live in a "nerf world," where everything is safe for everyone. Life is Risk and living out of our comfort zone.

I feel that it's important for each of us that is "taking a side," to understand that folks "on the other side" aren't necessarily enemies. They may be blinded by the personal imperative of the issue, but they aren't actually "out to get you."

The politicians may be two-faced creeps, using people's legitimate fears as leverage, but they aren't the rule.

For the record, I am vehemently against "Nerf World," but there's a very, very real problem here, and it would be good if we could approach it in a reasonable manner.

nickthegreek• 11 months ago

It could be a good idea to see if they do have ChatGPT and if so, it might be helpful to have the Customized GPT turned on with a note that they suffer from X and if signs are showing of Y, then it should attempt to Z.

dom96• 11 months ago

With the URL being https://futurism.com/openai-scanning-conversations-police.

This is actually the URL: https://futurism.com/man-chatgpt-psychosis-murders-mother

wslh• 11 months ago

I imagine now a lot of bots simulating madness to perform DoS.

akomtu• 11 months ago

It looks like what AI is doing to our society is it's substituting our moral compass to lead us off the cliff. A good deal of people today no longer know what's right and what's wrong, so they turn to an authority for advice. Turns out this authority is a chatbot who knows a lot, but has no moral compass whatsoever. Imagine someone goes to a pastor to resolve his moral dilemma, but the pastor behind the curtain happens to be a chatbot who only fakes morality. But that's what's happening now. The more people talk to this digital abomination, the more they will get derailed.

LamerBeeI9• 11 months ago

the chatbot knows a lot about morality - it's just not doing anything about it

jiveturkey• 11 months ago

AI psychosis

Is that in the DSM?

davsti4• 11 months ago

Not yet... but the machines will make it so ;)

causal2• 11 months ago

Yeah I dislike that term, because it puts the responsibility on the AI instead of the humans behind it. You wouldn't call it "paper psychosis" if OpenAI had written a letter encouraging psychotic behavior.

neilv• 11 months ago

The solution might be easy: recognize liability.

When a human cajoles someone into killing themselves, they probably get a criminal trial, as well as a civil lawsuit from the family.

Imagine a company set up an advice telephone line, and staffed it entirely with humans who were known clinical bullshitters. The company knows each call has a significant chance of the staffer simply making something up, or saying something absolutely crazy. The company markets the clinical bullshitters as geniuses, and somewhere there's the most mild disclaimer they could craft, about how the staffer might say something incorrect.[*]

Every time one of their staff persuades someone to kill themselves... that company is looking at criminal and civil liability.

The company absolutely knew this would happen, and that it will keep happening, but they are betting that they can keep getting away with this and other "externalities", and become millionaires or billionaires. Before legislators and courts realize that, just because the company inserted this new-fangled telephone technology into the interaction, that doesn't mean that existing laws don't apply to the interaction.

[*] Analogous to the tiny gray smallprint search engine LLM summary one that isn't even shown unless you click to see more, and then only after you scroll all the way to the bottom of that and look for it. Or the chat UI tiny gray smallprint that says you should "double-check" (no explanation what that means), and makes even this weak disclaimer not even text-selectable, so even if you see it, it's lost when you copy&paste the LLM response. Understated, and in what looks like bad faith. On the telephone in our example, it's a single tiny beep, at and the end of the call, that's actually the words "This may not be correct" sped up 3x, and whispered, and is often missed because the customer hangs up, but that's by design.*

duxup• 11 months ago

This tech came largely from the internet and just present a chat interface.

I'm not convinced this tech is more dangerous than what is already available, or that some of these folks wouldn't already do a thing ...

Mars008• 11 months ago

As a species we are so not ready for this tech.

Talk for yourself, please. Some humans, yes, can't be trusted even with a fork. There was even a movie about it. But extending it to the whole mankind is a bit of a stretch.

overfeed• 11 months ago

As a species we are so not ready for this tech.

On the contrary - it's the tech that's not ready for general availability.

Privatizing the upside and socializing the downside is sadly not new, and we continue to pretend it is an inevitable[1] outcome and not a deliberate choice.

1. https://news.ycombinator.com/item?id=44567857

dyauspitr• 11 months ago

We are ready. Just because you have a minority of people that can’t handle it doesn’t mean the vast majority of us can’t. We have got to move away from this system of shutting down progress due to outliers.

miletus• 11 months ago

This is exactly why a lot of people are running local LLMs or moving toward privacy-first platforms

we recently shipped secure mode on https://www.agentsea.com.

With Secure Mode, all chats run either on open-source models or models hosted on our own servers - so you can chat with AI without worrrying about privacy.

jonathanlb• 11 months ago

Given that this is in response to a ChatGPT user who killed his mother and then himself, I'm not sure that positioning your product as being more secure than ChatGPT is wise, because your marketing here suggests either:

1. Profound tone-deafness about appropriate contexts for privacy messaging

2. Intentional targeting of users who want to avoid safety interventions

3. A fundamental misunderstanding of your ethical obligations as an AI provider

None of these interpretations reflect well on AgentSea's judgment or values.

kbelder2• 11 months ago

I disagree. The fact that the crimes done by a mentally ill person are going to be used as a justification for surveillance on the wider population of users is a strong ethical reason to advocate for more security.

exe343• 11 months ago

Are you in America? Do you also support banning guns?

sleazebreeze2• 11 months ago

Or maybe I just want to be able to talk to an LLM without worrying about if its going to report me to the authorities.

kbelder2• 11 months ago

There is a very strong use case for less-powerful but local LLMs, and there's going to be a big expansion in that area in the next couple years. So big, I'll bet that all the major AI players will do everything they can to cripple them.

brazukadev• 11 months ago

Not Nvidia, that's their best scenario

conradkay• 11 months ago

Unfortunately I think you're overestimating how many people care enough about privacy to go through the effort of running LLMs locally and likely buying a GPU

lawlessone2• 11 months ago

tbh i just run local because i can . No real reason too.

edit: there are some instances where i would like to be able to set the same seed repeatedly which isn't always possible online.

datameta3• 11 months ago

Realistically, how useful is local LLM usage? What are your use cases, hardware, and models used?

themafia• 11 months ago

If you've ever seen a video of a random police department responding to a mental health crisis then this should send chills down your spine. They are not equipped to handle this type of reporting. They are a cudgel and not a scalpel.

throwaway2903• 11 months ago

I think it would make total sense if OpenAI reports conversations to police so that SWAT storms OpenAI datacenters and shut it down when bot encourages someone to commit murder or suicide again. But something tells me it's probably not that

subscribed2• 11 months ago

No, that would be shooting dead the person accused of having suicidal thoughts.

itqwertz• 11 months ago

I’ve had a few conversations with people who use ChatGPT as a therapist. There is real danger of using LLMs that are engineered agree with you, at least in terms of therapy.

miltonlost2• 11 months ago

Human therapists? Years of training, certification, can be de-licensed.

AI Therapists? No rules! Do what you want! Move fast, break people! (Except in Illinois. They fortunately banned AI use in therapy)

ACCount375• 11 months ago

It doesn't matter.

Even if "AI cannot legally do therapy", nothing would stop people from crying their soul out to a "Gangster Sponge Bob" persona on some "funny novelty chatbot" website.

What do you do about that?

nakamoto_damacy2• 11 months ago

LLMs have no embodied sense of the world. They talk as if they understand and experience emotion, when they don't have actual embodied feelings/sensations/pain/pleasure/etc.

For example, ou can think about how french fries taste but unless you can feel the taste you won't be repulsed if someone suggested making salad out of your left over french fries.

Here is an actual chat with GPT5, that replays a experiment from the most recent South Park episode:

Me: I want to build and market a contraption that turns left over french fries into salad

ChatGPT said:

That’s a creative idea! Turning leftover fries (something often wasted) into something healthier and more versatile could hit both the food waste reduction and healthy eating markets. Let’s break it down:

1. Concept Feasibility

Core Problem Solved: Leftover fries usually go soggy, stale, and unappealing. You want to “upcycle” them.

Potential Process:

A kitchen gadget that breaks fries down into smaller bits (like croutons, or a potato base).

Mixing with fresh greens, herbs, dressing, or flavor infusions.

Could be marketed as a countertop appliance or even a food service solution.

2. Target Markets

Households: Especially families, students, or anyone who regularly orders fries and hates waste.

Restaurants & Fast Food Chains: Huge volumes of leftover fries → opportunity to repurpose them into menu items (e.g., “French Fry Salad Bowls”).

Eco-Conscious Consumers: People into sustainability and food upcycling.

3. Product Format Ideas

Appliance: Like a mini food processor that’s specialized for fries → chops, seasons, mixes into salad.

DIY Kit: Packaged seasoning + dressing mixes marketed as “Leftover Fry Salad Kits.”

B2B Solution: Larger-scale machines for restaurants.

4. Challenges

Texture: Fries lose crunch; you’ll need to re-crisp or season to make them appealing in a salad.

Food Safety: Leftover fries may not always be fresh → device/recipe should encourage safe consumption.

Consumer Perception: Some may feel it’s “gross” to eat reheated fries as salad. Marketing needs to make it fun, eco-friendly, and tasty.

5. Marketing Angles

Sustainability: “Save your fries, save the planet.”

Health Twist: “Turn guilty pleasure into a healthy bite.”

Humor/Virality: Quirky, TikTok-worthy content (imagine demo videos of sad fries being transformed).

Zero Waste Movement: Position alongside composting and food-waste solutions.

6. Next Steps

Prototype: Could start small by testing recipes (turn fries into croutons, hash-style toppings, etc.).

Research: Do people actually want this? Surveys or small focus groups can test appetite.

Branding: Quirky, fun name (e.g., “FrieFresh,” “SpudCycle,” “Fry2Fit”).

Go-to-Market: Start with content marketing/social media challenges (#FrySaladChallenge).

Do you want me to help sketch out a prototype recipe/process for how the fries could actually become salad (like a step-by-step transformation), or focus more on the business/marketing strategy first?

exe343• 11 months ago

#shutupandtakemymoney

(if you knew how most processed food were made, you wouldn't touch anything other than fresh vegetables).

cyberax2• 11 months ago

If you knew how much (literal) shit is on fresh vegetables, you'd switch to using IV sugar solutions.

dns_snek• 11 months ago

All food (other than raw produce) is processed. Cooking meat over a fire makes it processed food. You'll have to be more specific.

Kuinox2• 11 months ago

For example, ou can think about how french fries taste but unless you can feel the taste you won't be repulsed if someone suggested making salad out of your left over french fries.

Deep fried potato salad is a thing and humans didnt waited LLMs to invent it.

nakamoto_damacy2• 11 months ago

[Yuk,] a thing where?

pton_xd• 11 months ago

Anthropic is also scanning conversations and reporting them to the authorities [0]. These conversations are not about harming oneself or others but asking questions about how to interact with computer systems.

[0] https://www.anthropic.com/news/detecting-countering-misuse-a...

fakedang• 11 months ago

I am a North Korean agent apparently. Hopefully me calling the AI a dipshit or a dumbass at regular intervals puts me in the clear.

SilverElfin• 11 months ago

Unacceptable. This type of behavior can be used for an increasing expansive regime of censorship or worse. Imagine being turned into authorities for, say, discussing ways to help asylum applicants or under a different administration, the lab leak theory, or under a different administration, how to grow cannabis. This is why the ACLU of the past had it right, when they were defending even offensive uses of the right to free speech. And to be able to speak freely, you need anonymity and privacy.

tuatoru• 11 months ago

Private companies can do as they choose within the law.

You can choose not to use their products.

subscribed2• 11 months ago

You may not know if you're using the product.

Miraste• 11 months ago

Whatever private companies you route your communications through are almost certainly working on implementing this. The NSA will be applying this technology to ~all communication routed through the USA, as well. You can't opt out of it.

johnfn• 11 months ago

Didn't we have a conversation a few days ago where a kid committed suicide while documenting it on ChatGPT and the clear consensus was that if OpenAI had been scanning user's conversations and reporting them to the police, this could have been averted, and this was a huge issue? But now when they go and do it, everyone says this is completely unacceptable? What exactly is OpenAI supposed to do, then?

miltonlost2• 11 months ago

Frankly not exist in its current form

mjamesaustin• 11 months ago

Don't release a model that's so sycophantic it provides encouraging advise to suicidal users about how and where their body should be discovered after they kill themselves.

Instead, it should warn the user that their pattern of thinking indicates mental health issues and they need to search for health from a medical professional.

nomel4• 11 months ago

need to search for health from a medical professional.

In all of the cases (that I'm aware of), these people were already diagnosed, and the family was aware of their mental illnesses. The awareness and agency you're asking for, from the mentally ill individual, would probably preclude their condition.

If someone is mentally ill, sometimes the only practical protection is to deem them unfit for access by the caregivers or the government (as we do with guns).

The alternative is to assume everyone is unfit, which has its own practical and ethical problems.

cherioo• 11 months ago

Last week the think of the children groups were out in force, and this week it’s the privacy and liberty groups. Internet opinion is not a monoculture.

What does bug me though is that it was politically incorrect for the privacy folks to come out last week and for the two groups to fight. I have no solution.

fwip2• 11 months ago

I think this might be an instance of the Goomba fallacy: https://englishinprogress.net/wp-content/uploads/2025/04/ima...

righthand• 11 months ago

Wasn’t OpenAI making a big stink about NYT getting access to data? Here they’ve been feeding it to the cops the whole time.

LamerBeeI9• 11 months ago

Wouldn't say they have the whole time since this is a reaction to having an unsafe product that will tell you when, how, why and where to kill people.

bigmattystyles2• 11 months ago

At what point would an llm be considered to be practicing medicine without a license legally; or would that only apply to persons?

busymom0• 11 months ago

It's probably the parent corporation who would be liable for practicing medicine without a license and not the LLM (which is merely a software created by the said corporation).

bigmattystyles2• 11 months ago

Right, I agree, but my question stands - isn’t that what they are doing?

blindriver• 11 months ago

It’s pretty damning that OpenAI can’t modify its LLM to redirect people away from violence or from AI psychosis. Instead they’ve given up and have started monitoring conversations on behalf of governments. Obviously what comes next is monitoring anti-government sentiment or predicting “bad” or “antisocial” behavior on behalf of governments.

sho_hn7• 11 months ago

What I would genuinely like to see in ChatGPT and similar services are well-implemented parental controls that allow a parent to monitor a child's conversations and get reports on them.

echelon_musk4• 11 months ago

Would you inform your children they were being surveilled?

sho_hn7• 11 months ago

Yes, I think parental controls should generally be transparent to children.

That said: Let me be clear that I'm very happy I'm father to a two-year old and we have some time to figure out our "AI policy" still, and for the tech/services to improve. I don't envy parents to the 8+ crowd currently.

thatguymike• 11 months ago

Good news! It's being worked on, announced in the blog post linked from that article. https://openai.com/index/helping-people-when-they-need-it-mo...

user2144124122• 11 months ago

the 'ai safety' was and is all along CENSORSHIP. Now there's a surveillance element on top. Truly horrific.

LamerBeeI9• 11 months ago

on one hand there are dead people, and then you have "my rights". Truly horrific.

Argonaut998• 11 months ago

Yes, this argument is also used against the US’ second amendment. The rights of everyone are more important than a few dead mentally ill people.

Esophagus44• 11 months ago

On one hand, the media hammered OpenAI for not doing enough to protect people from the possible harms of AI.

Then OpenAI takes steps to mitigate those risks, and it becomes a salacious story as well.

Society having no idea what to expect from these tools makes this a difficult situation for LLM providers.

ep103• 11 months ago

It wouldn't be a difficult situation if these guys were ethical shops from the get-go, but they aren't, they're trying to staple minimally required ethics on afterwards, and it shows.

Ajedi32• 11 months ago

LLMs are not moral agents. Any attempt to make them behave that way will necessarily be "staple[ed] on afterwards".

Esophagus44• 11 months ago

To play devil’s advocate, what ethical safeguards are OpenAI responsible for that they have failed to implement?

This is a wild and difficult to understand technology, even for the people building it. And their safeguards are constantly evolving.

I think you’re attributing to malice what should be attributed to people commercializing a novel technology that is, frankly, being exploited by users.

oceanplexian• 11 months ago

On one hand, the media hammered OpenAI for not doing enough

How about ignore the increasingly irrelevant mainstream media and do what you feel is right?

Esophagus44• 11 months ago

Huh?

You watch too much TV - there are plenty of decent news sources that publish concerned pieces about OpenAI’s effects on society at large. And it is obviously having an effect on OpenAI as well.

user2144124122• 11 months ago

yes lets auto dispatch killer cops to people having a mental health crisis. How streamlined.

teeray• 11 months ago

The new swatting will probably be malware that stealthily uses another LLM to convince ChatGPT / Claude that you're trying to commit a crime. Unless, of course, you pay a ransom.

causal2• 11 months ago

Prompt-injected SWATing even.

senko3• 11 months ago

The right way to deal with this is education.

1. We should be educating the populace about failure modes of AI chatbots - something well known to HNers but not to the general public.

2. CEOs, pundits, and marketers should be held accountable (morally, in the court of public opinion and in the news) when they present AI as unvarnished good - that's false marketing at the very least, and leads to tragic consequences.

3. The cat's out of the bag. With so many AI providers and models to choose from, and trivially-ran local models, you can't possibly police them all.

4. Instead of trying to shield the users (a futile task), we should educate them and equip them with knowledge on how to safely use the tools.

5. I'm cynical enough to believe governments all over the world will use the "think of the children" to institute de facto and/or de jure censorship and spying, pointing to the inability of the industry to treat the matter with actual respect.

pants22• 11 months ago

What do you envision in the form of education? "Pass this test to use ChatGPT"?

senko3• 11 months ago

As in informing the public, not adding to the school curriculum.

A sustained collaborative effort by the AI makers, (local) governments and journalists could work, IMHO.

Instead AI makers hype it up, governments have no clue wtf is going on and media just chases clicks.

johncolanduoni2• 11 months ago

I don’t think education is going to help much when a person is experiencing full-blown psychosis. It’s not as if mental health professionals don’t already try to give their patients tools to counter delusional thought patterns, there is just a limit to what knowledge of one’s condition can do once their thinking becomes that disordered.

senko3• 11 months ago

I agree, but it might prevent the person entering that state due to encouragement from the AI?

We don't require licenses for knives because a psychotic person may injure themselves or others.

echelon_musk4• 11 months ago

Educated consumers are harder to exploit and to convince to consume products that they might not need or want. It's antithetical to business and profit.

I'm cynical too.

Most peoples use of these tools is predicated in not knowing their flaws. If they did, I wonder if they would use the products as much?

drnick1• 11 months ago

One more reason to run your LLM locally, just like everything else (file servers, email, Matrix channel, Home Assistant...)

LamerBeeI9• 11 months ago

I agree, nothing should get in the way of your murder suicide plots.

aktuel• 11 months ago

chatgpt is barely usable at all anyhow atm. The amount of complete garbage responses gpt-5 is giving me is mind-boggling. just today it ran into an infinite loop of endless printout again by a completely innocuous query. I honestly cannot understand how its stock is still where it's at.

LamerBeeI9• 11 months ago

what stock?

djoldman• 11 months ago

Escalate risk of physical harm to others for human review. When we detect users who are planning to harm others, we route their conversations to specialized pipelines where they are reviewed by a small team trained on our usage policies and who are authorized to take action, including banning accounts. If human reviewers determine that a case involves an imminent threat of serious physical harm to others, we may refer it to law enforcement. We are currently not referring self-harm cases to law enforcement to respect people’s privacy given the uniquely private nature of ChatGPT interactions.

https://openai.com/index/helping-people-when-they-need-it-mo...

mallowdram3• 11 months ago

All AI does is arbitrarily speed up indirectness in arbitrary signals by predicting a next word. Course-correcting the arbitrary never reaches directness or specifics. This is the expected outcome of metaphors (words) pretending to have meaning without context. It's abysmal how this came to market.

The overall lack of depth in CS academic programs and research is on full display here. Anyone in Systemic Functional Linguistics can diagnose this as the embedded social dark matter of language: domination, control, status stuck inside the models like a bizarre daemon.

Ultimately language is irrelevant, it has to be replaced. That is the only function AI generally reveals.

insane_dreamer4• 11 months ago

Minority Report got it wrong. The pre-cogs weren't humans; they were LLMs.

insane_dreamer4• 11 months ago

Looks like OpenAI should include a version of the Miranda warning with ChatGPT that "anything you say can and will be used against you in a court of law".

msgodel• 11 months ago

Every online service is compelled by law to answer subpoenas. You're supposed to assume the collaborate with the state.

insane_dreamer4• 11 months ago

OpenAI isn't talking about responding to subpoenas; it's volunteering information to the gov. That's a very different thing.

furyofantares• 11 months ago

Everything about this sucks. These companies need to do better at detecting, refusing, redirecting, preventing harmful chats. They need to offer this to anyone using the APIs to build products too.

And that all also sucks. I don't trust these companies one bit to be monitoring all of these. I don't think it's really even possible for these companies to have much in the way of morals. So they also need to NOT do any of that.

And then there's the issue of reporting to authorities. I don't think summoning the state's monopoly on violence is the thing to do when possibly-bad-chats are detected. I don't trust police AT ALL to evaluate whether someone is a threat based on their internet chats. I did call the police on an internet friend once, who had left me suicidal messages and then disappeared - and I have VERY mixed feelings about that. I didn't know any other way to get someone to try to get to him. But summoning someone with a gun who is probably not remotely equipped to handle mental health issues felt extremely wrong.

Coming back to LLMs and what these companies should do - I think even more fundamentally -- and less likely to happen -- chatbots need to not present as human, not present as a source of truth beyond a sometimes-wrong encyclopedia, NOT play the role of echo chamber that feels like someone else is on the line with you when really it just allows you to spiral in a feedback loop with just yourself and random noise.

I love this technology and yet I am tempted to say, shut it all down. Of course, that won't happen. But it is how I feel at times.

ChrisArchitect• 11 months ago

[dupe]

This story didn't just break.

More discussion: https://news.ycombinator.com/item?id=45088651

WSJ source last week: https://news.ycombinator.com/item?id=45059030

OpenAI's response from a week ago: https://news.ycombinator.com/item?id=45050635

In Search of AI Psychosis

https://news.ycombinator.com/item?id=45027072

heisenbit• 11 months ago

Considering how heavy handed the police in the US, particulary these days is action this is a scary throught. ChatGPT 4.1 a few days ago set me up as a black hat hacker - first it delivered subpar code and then when I asked it what would happen if the trailing / was removed it accused me of hacking (well, it told me that I was asking a question it is not anwering). It set a trap and I triggered it. Now this was most likely not reported to police or not likely to the company who's enterprise GPT I was using.

Still it is a scary thought. When working with an LLM we may be playing around and we here includes the LLM playing with us. If this starts having real world consequences without anyone having done anything we are close to pre-crime enforcement.

normalaccess• 11 months ago

Great, replace search with a new "tool" that is fully capable of autonomously analyzing user behavior and hand that data over on a silver platter to law enforcement bypassing all the legal red tape.

AI truly is a Faustian Bargain. A powerful daemon in your pocket that can do wonderous things, but is still a daemon.

joshcsimmons• 11 months ago

The company provides a probabalistic predictive text "chat". The responsibility ends there. I pay for electricity but Edison Electric doesn't ring the police when I get close to the power outlet with a metal fork.

This opens to door to a ton of creepy things.

I_dream_of_Geni2• 11 months ago

Oh, I'm SURE this is gonna work out well......

smsm422• 11 months ago

I was on the fence whether to keep my subscription, given there are many alternatives and really I'm not using it that much anymore. Now, given how easily they just decide they would call the police on me if they don't like something, I guess that decides it.

OpenAI says it's scanning users' conversations and reporting content to police

Related Stories