Not-so-fun facts: Clearview AI was founded by white supremacists [1] and some of its first desired users were white supremacists and authoritarians [2][3]. Peter Thiel backed the company and it was able to get through a round of lawsuits over its privacy- and TOS-breaking scraping used to populate its facial recognition database.
Always easier when you can avoid the law and just buy it off the shelf. It’s fine to do this, we say, because it’s not being done by the government - but if they’re allowed to turn around and buy it we’re much worse off.
That's why it doesn't make sense to ban governments from doing things while still allowing private companies. Either it is illegal to surveil the public for everyone, or the government can always do it indirectly with the same effect.
I don't think the deal described here is even that egregious. It's basically a labeled data scrape. Any entity capable of training these LLMs are able to do this.
The difference is that a government can take personal liberty away from people in the most direct way. A private company can't decide to lock somebody away in prison or send them to death row. (Hopefully anyway.) So we put a higher standard on government.
That said, I do believe there ought to be more restrictions on private use of these technologies.
I also personally think there are some private collections we should ban, or put in place limitations on how it can be used, in the interest of general privacy.
That is trickier to decide on and surely there's room to debate.
Yeah but these companies are operating hand in glove with govt such that there's no discernible difference between the current system and government just doing it themselves. Ban it outright.
I don't disagree with the sentiment. I feel like what we're seeing lately is that private companies are doing the thing that would violate the 4th amendment if government did it, then they sell to the government. The idea that it's not the government itself violating the constitution because they did it through a contractor is pretty absurd.
What specific legal measures you do to enforce this, I don't know, there's some room for debate there.
I don't think there is an expectation of privacy for things you literally post to the public, like social media. Even the government doing the scraping directly I believe would not violate the 4th amendment.
The third party doctrine also basically legalizes most types of search through people's "cloud data". To have an expectation of privacy, the data needs to not be shared in the first place.
I don't think tying the hands of the government is a viable solution. The sensitive data needs to not be collected in the first place via technical and social solutions, as well as legislation to impose costs on data collection.
- Teaching that "the cloud is just someone else's computer"
- E2EE cloud
- Some way of sharing things that don't involve pushing them to the whole internet, like Signal's stories.
- GDPR type legislation which allows deleting, opting out, etc
But that is his point with "or the government can always do it indirectly with the same effect"
The company doesn't have that power, but the government can compel companies to provide them with the same data as long as it exists, and then abuse it in the same way as if they had collected it themselves.
The separation between private and the government is purely theatrics - a mere administrative shell.
I really don't understand why people treat it with such sacrosanct reverence.
It reminds me of a cup and ball street scam. Opportunistic people move things around and there's a choir of true believers who think there's some sacred principles of separation to uphold as they defend the ornamental labels as if they're some divine decree.
In some cases yes, especially when it comes to surveillance, the distinction doesn't feel like very much. When the government hires a contractor specifically because they break the spirit of the 4th amendment, it's hard to argue that it's not the government breaking the law.
A private company can rat you out the government in the same way that a private citizen can report you to the police. I don't see a reasonable way to change this.
The government should be held to higher standards in terms of being able to appeal its actions, fairness, evidentiary standards. But the government shouldn't necessarily be prevented from acquiring and using information (which is otherwise legally obtained).
I don't disagree that we should perhaps more restrictions on private processing of data though -- GDPR style legislation that imposes a cost on data collection is probably sufficient.
This is why we should shun the people that build this stuff. If you take a paycheck to enable fascism, you're a bad person and should be unwelcome in polite society.
There are certain people who believe that average citizens can be held responsible for the actions of their government, to the point that they are valid military targets.
Well, if that's true then employees of the companies that build the tools for all this to happen can also be held responsible, no?
I'm actually an optimist and believe there will come a time whena whole lot of people will deny ever working for Palantir, for Clearview on this and so on.
What you, as a software engineer, help build has an impact on the world. These things couldn't exist if people didn't create and maintain them. I really hope people who work at these companies consider what they're helping to accomplish.
I never worked at a company that could broadly be considered unethical, I don't think. But it was always a bit disheartening how many little obviously unethical decisions (e.g., advertised monthly plans with a small print "annual contract" and cancellation fee) almost every other employee would just go along with implementing, no pushback whatsoever. I don't know what it is, but your average employee seemingly sees themselves as wholly separate from the work they're paid to do.
I have friends who are otherwise extremely progressive people, who I think are genuinely good people, who worked for Palantir for many years. The cognitive dissonance they must've dealt with...
We need a Constitutional amendment that guarantees a complete right to anonymity at every level: financial, vehicular, travel, etc. This means the government must not take any steps to identify a person or link databases identifying people until there has been a documented crime where the person is a suspect.
Only if an anonymous person or their property is caught in a criminal act may the respective identity be investigated. This should be sufficient to ensure justice. Moreover, the evidence corresponding to the criminal act must be subject to a post-hoc judicial review for the justifiability of the conducted investigation.
Unfortunately for us, the day we stopped updating the Constitution is the day it all started going downhill.
Maybe. Anonymity is where bad actors play. Better to have better disclosure and de-anonymization in some cases. If some live in fear (e.g. of cartels), go after the cartels harder than they go after you.
Anonymity is where little bad actors play. The big ones don't need to be anonymous because their nefariousness is legal, or they don't get prosecuted. See: waves vaguely in the direction of the US government.
That said, the recent waves vaguely in the direction of the US government has demonstrated the weakness of legal restrictions on the government. It's good to have something you can point to when they violate it, but it's too easily ignored. There's no substitute for good governance.
That is a myth spread by control freaks and power seekers. Yes, bad actors prefer anonymity, but the quoted statement is intended to mislead and deceive because good actors can also prefer strong anonymity. These good actors probably even outnumber bad ones by 10:1. To turn it around, deanonymization is where the bad actors play.
Also, anonymity can be nuanced. For example, vehicles can still have license plates, but the government would be banned from tracking them in any way until a crime has been committed by a vehicle.
Not sure why you say that statement was intended to deceive?
Both good and bad actors benefit in the current system from anonymity. If bad actors had their identities revealed, they'd have a lot harder time being a bad actor. Good actors need anonymity because of those bad actors.
That will be wildly unpopular with both parties and most importantly their constituents. I doubt even the libertarian party should they get the president, house and senate could pull it off
Note that the Amendment would apply only to the government, not to private interests. Even so, i could be unpopular among advertisers and data resellers, e.g. Clearview, who sell to the government. I guess these are what qualify as constituents these days. The people themselves have long been forgotten as being constituents.
What do you mean "even" the libertarian party? Libertarians would remove whatever existing laws there are around facial recognition so that companies are free to do whatever they like with the data.
Don't we already have facial recognition technology that isn't based on AI? why is throwing AI into the mix suddenly a reasonable product? Liability wavers?
I think the facial rec systems you're thinking of will recognize faces, but not ID them. They need you to label a face, and then it recognizes that face with a name from there on. Clearview is different in that you can provide it an unknown face and it returns a name. Whether it's just some ML based AI vs an LLM, it's still under the AI umbrella technically.
There are plenty of facial rec systems. Thinking of systems like in iOS Photos, or any of the other similar photo library systems. I think pretty much everyone would be freaked out if they started IDing people in your local libraries.
Note that there is no difference in the model or in the training. The only thing needed to convert ios photos into one that IDs people is access to a database mapping name to image. The IDing part is done after the "AI" part, it's just a dot product.
unsure what you mean by starting IDing? Majority business in US does it already, all banks use facial recognition to know who comes through their door (friend who works in IT at Bank of America told me they implemented it cross all Florida branches sometime in 2009), most large chain gas stations as well, so does car rentals, most hotels, etc. I was recently booted out of Mazda Dealership in Florida because 11 years ago in Georgia I sued Toyota Dealership for a lemon sell, and now they both under same ownership and my name came up on "no business" alert when I entered their offices.
After the literal first one which just measured distance between nose and mouth and stuff like that from the 1960s, everything else has been based on AI.
If my memory serves me, we had a PCA and LDA based one in the 90s and then the 2000s we had a lot of hand-woven adaboosts and (non AI)SIFTs. This is where 3D sensors proved useful, and is the basis for all scifi potrayals of facial recognition(a surface depth map drawn on the face).
In the 2010s, when deep learning became feasible, facial recognition as well as all other AI started using an end to end neural network. This is what is used to this day. It is the first iteration pretty much to work flawlessly regardless of lighting, angle and what not. [1]
Note about the terms AI, ML, Signal processing:
In any given era:
- whatever data-fitting/function approximation method is the latest one is typically called AI.
- the previous generation one is called ML
- the really old now boring ones are called signal processing
Sometimes the calling-it-ML stage is skipped.
[1] All data fitting methods are only as good as the data. Most of these were trained on caucasian people initially so many of them were not as good for other people. These days the ones deployed by Google photos and stuff of course works for other races as well, but many models don't.
And this right here is why Clearview (and others) should have been torn apart back when they first appeared on stage.
I 'member people who warned about something like this having the potential to be abused for/by the government, we were ridiculed at best, and look where we are now, a couple of years later.
I think anything short of fully obscuring your face (a-la ICE-agent/stormtrooper) will be merely a mitigation and not 100% successful. I recall articles talking about face recognition being used "successfully" on people wearing surgical masks in China. In the US they ask you to remove face masks in places where face recognition is used (at the border, TSA checkpoints), but would be unsurprised if that isn't strictly needed in most cases (but asking people to remove it preemptively ends up being faster for throughput).
99.9% of people walk around with an electronic device that identifies them. If a particular person doesn’t, it should be trivial to filter out all the people that it couldn’t have been, leaving only a small list of possible people.
(ProPublica, 404media, APM Marketplace, Associated Press, Vox, Block Club Chicago, Climate Town, Tampa Bay Times, etc get my journalism dollars as well)
Having AI assisted law enforcement will be a big force of making the law applied evenly. Law enforcement has limited resources so being able to give them a force multiplier will help clean up a lot of issues that were thought to be impossible to enforce before.
This is exactly, precisely the opposite of what the impact will be.
For example:
- every technology has false positives. False positives here will mean 4th amendment violations and will add an undue burden on people who share physical characteristics with those in the training data. (This is the updated "fits the description."
- this technology will predictably be used to enable dragnets in particular areas. Those areas will not necessarily be chosen on any rational basis.
- this is all predictable because we have watched the War on Drugs for 3 generations. We have all seen how it was a tactical militaristic problem in cities and became a health concern/addiction issues problem when enforced in rural areas. There is approximately zero chance this technology becomes the first use of law enforcement that applies laws evenly.
Not only is this incredibly naive, it misses that whole "consent of the governed" thing. I don't want AI involved in policing. They are bad enough and have so little accountability without "computer says so" to fall back on, That's all AI will do, make a bad situation worse.
The targets for the AI are still set by humans, the data the AI was trained on is still created by humans. Involving a computer in the system doesn't magically make it less biased.
are you sure it won't enabled targeted enforcement for people law enforcement finds irritating, more than evenly applied law? It's still people setting the priorities and exercising discretion about charging.
do you think the records of the vast number of police departments and agencies would be combinable with the separate court records, as well as the facial recognition access data source (if it exists?)
Then cops should be taken out of the core law enforcement agentic loop. There could be a new role of people who the AI dispatches instead to do law enforcement work in the real world.
I think you fundamentally misunderstand what the role of the police is. They protect property, the owning class, and the status quo. Laws are just a tool for them to do that. Equal justice for all is not a goal for them, and AI will not provide more of it.
The thing is if you have a truly fair AI you start catching the Trumps and Musks of this world in their little underaged trists. How long do you think that system would actually stay running for?
The thing you're missing is our system is working exactly like it's supposed to for rich people.
Why do you write so many low-effort, disingenuous, inflammatory comments? They're "not even wrong", yet they just suck energy right out of productive discussion as people inevitably respond to one part of your broken framing, and then they're off to the races arguing about nonsense.
The main problem with the law not being applied evenly is structural - how do you get the people tasked with enforcing the law to enforce the law against their own ingroup? "AI" and the surveillance society will not solve this, rather they are making it ten times worse.
I want to share my opinion even if I know that it may not be a popular one on HN. I am not trying to maximize my reputation by always posting what I believe will get the most upvotes, but instead I prioritize sharing my opinion.
>people inevitably respond to one part of your broken framing, and then they're off to the races arguing about nonsense.
I agree that this unproductive. When people have two very different viewpoints it is hard for that gap to be bridged. I don't want to lay out my entire world view and argument from fist principals because it would take too much time and I doubt anyone would read it. Call it low effort if you want, but at least discussions don't turn into a collection of a single belief.
>how do you get the people tasked with enforcing the law to enforce the law against their own ingroup?
Ultimately law enforcement is responsible to the people so if the people don't want it then it will be hard to change. In regards to avoiding ingroup preference it would be worth coming up with ways of auditing cases that are not being looked into and having AI try to find patterns in what is causing it. The summaries of these patterns could be made public to allow voters and other officals to react to such information and apply needed changes to the system.
[1] https://www.huffingtonpost.co.uk/entry/clearview-ai-facial-r...
[2] https://www.buzzfeednews.com/article/carolinehaskins1/clearv...
[3] https://www.nytimes.com/2020/01/18/technology/clearview-priv...
[1] https://img.huffingtonpost.com/asset/5e8cc7922300005600169bd...
I don't think the deal described here is even that egregious. It's basically a labeled data scrape. Any entity capable of training these LLMs are able to do this.
That said, I do believe there ought to be more restrictions on private use of these technologies.
A private company can 100% do this in many ways. They already do this buy putting up and using their technology in minority areas, for example.
We should ban the government from accessing data gathered by private companies by default, perhaps. I need to mull on it.
That is trickier to decide on and surely there's room to debate.
What specific legal measures you do to enforce this, I don't know, there's some room for debate there.
I don't think tying the hands of the government is a viable solution. The sensitive data needs to not be collected in the first place via technical and social solutions, as well as legislation to impose costs on data collection.
- Teaching that "the cloud is just someone else's computer"
- E2EE cloud
- Some way of sharing things that don't involve pushing them to the whole internet, like Signal's stories.
- GDPR type legislation which allows deleting, opting out, etc
The company doesn't have that power, but the government can compel companies to provide them with the same data as long as it exists, and then abuse it in the same way as if they had collected it themselves.
I really don't understand why people treat it with such sacrosanct reverence.
It reminds me of a cup and ball street scam. Opportunistic people move things around and there's a choir of true believers who think there's some sacred principles of separation to uphold as they defend the ornamental labels as if they're some divine decree.
I mean come on. Know when you're getting played.
The government should be held to higher standards in terms of being able to appeal its actions, fairness, evidentiary standards. But the government shouldn't necessarily be prevented from acquiring and using information (which is otherwise legally obtained).
I don't disagree that we should perhaps more restrictions on private processing of data though -- GDPR style legislation that imposes a cost on data collection is probably sufficient.
[^1]: https://arstechnica.com/tech-policy/2025/09/court-rejects-ve...
Well, if that's true then employees of the companies that build the tools for all this to happen can also be held responsible, no?
I'm actually an optimist and believe there will come a time whena whole lot of people will deny ever working for Palantir, for Clearview on this and so on.
What you, as a software engineer, help build has an impact on the world. These things couldn't exist if people didn't create and maintain them. I really hope people who work at these companies consider what they're helping to accomplish.
I have friends who are otherwise extremely progressive people, who I think are genuinely good people, who worked for Palantir for many years. The cognitive dissonance they must've dealt with...
Hannah Arendt coined the term “the banality of evil”. Many people think they are just following orders without reflecting on their actions.
Only if an anonymous person or their property is caught in a criminal act may the respective identity be investigated. This should be sufficient to ensure justice. Moreover, the evidence corresponding to the criminal act must be subject to a post-hoc judicial review for the justifiability of the conducted investigation.
Unfortunately for us, the day we stopped updating the Constitution is the day it all started going downhill.
That said, the recent waves vaguely in the direction of the US government has demonstrated the weakness of legal restrictions on the government. It's good to have something you can point to when they violate it, but it's too easily ignored. There's no substitute for good governance.
That is a myth spread by control freaks and power seekers. Yes, bad actors prefer anonymity, but the quoted statement is intended to mislead and deceive because good actors can also prefer strong anonymity. These good actors probably even outnumber bad ones by 10:1. To turn it around, deanonymization is where the bad actors play.
Also, anonymity can be nuanced. For example, vehicles can still have license plates, but the government would be banned from tracking them in any way until a crime has been committed by a vehicle.
Both good and bad actors benefit in the current system from anonymity. If bad actors had their identities revealed, they'd have a lot harder time being a bad actor. Good actors need anonymity because of those bad actors.
If my memory serves me, we had a PCA and LDA based one in the 90s and then the 2000s we had a lot of hand-woven adaboosts and (non AI)SIFTs. This is where 3D sensors proved useful, and is the basis for all scifi potrayals of facial recognition(a surface depth map drawn on the face).
In the 2010s, when deep learning became feasible, facial recognition as well as all other AI started using an end to end neural network. This is what is used to this day. It is the first iteration pretty much to work flawlessly regardless of lighting, angle and what not. [1]
Note about the terms AI, ML, Signal processing:
In any given era:
- whatever data-fitting/function approximation method is the latest one is typically called AI.
- the previous generation one is called ML
- the really old now boring ones are called signal processing
Sometimes the calling-it-ML stage is skipped.
[1] All data fitting methods are only as good as the data. Most of these were trained on caucasian people initially so many of them were not as good for other people. These days the ones deployed by Google photos and stuff of course works for other races as well, but many models don't.
I 'member people who warned about something like this having the potential to be abused for/by the government, we were ridiculed at best, and look where we are now, a couple of years later.
https://www.law.cornell.edu/uscode/text/18/1001
I don't think I've read a Wired article since 2002...
Most Americans don’t pay for news and don’t think they need to - https://news.ycombinator.com/item?id=46982633 - February 2026
(ProPublica, 404media, APM Marketplace, Associated Press, Vox, Block Club Chicago, Climate Town, Tampa Bay Times, etc get my journalism dollars as well)
For example:
- every technology has false positives. False positives here will mean 4th amendment violations and will add an undue burden on people who share physical characteristics with those in the training data. (This is the updated "fits the description."
- this technology will predictably be used to enable dragnets in particular areas. Those areas will not necessarily be chosen on any rational basis.
- this is all predictable because we have watched the War on Drugs for 3 generations. We have all seen how it was a tactical militaristic problem in cities and became a health concern/addiction issues problem when enforced in rural areas. There is approximately zero chance this technology becomes the first use of law enforcement that applies laws evenly.
I think that is pretty unlikely
Crimes aren't solved, despite having a literal panopticon. This view is just false.
Cops are choosing to not do their job. Giving them free access to all private information hasn't fixed that.
The thing you're missing is our system is working exactly like it's supposed to for rich people.
Go lick boots elsewhere.
The main problem with the law not being applied evenly is structural - how do you get the people tasked with enforcing the law to enforce the law against their own ingroup? "AI" and the surveillance society will not solve this, rather they are making it ten times worse.
>people inevitably respond to one part of your broken framing, and then they're off to the races arguing about nonsense.
I agree that this unproductive. When people have two very different viewpoints it is hard for that gap to be bridged. I don't want to lay out my entire world view and argument from fist principals because it would take too much time and I doubt anyone would read it. Call it low effort if you want, but at least discussions don't turn into a collection of a single belief.
>how do you get the people tasked with enforcing the law to enforce the law against their own ingroup?
Ultimately law enforcement is responsible to the people so if the people don't want it then it will be hard to change. In regards to avoiding ingroup preference it would be worth coming up with ways of auditing cases that are not being looked into and having AI try to find patterns in what is causing it. The summaries of these patterns could be made public to allow voters and other officals to react to such information and apply needed changes to the system.