Just like breaking crypto in the age of cloud is more about cost than time, this will lead to legal attacks based on the same principle. The biggest wallet wins.
Will these models eventually replace all knowledge work, leaving lawyers, doctors, product managers, software developers, and others out of a job?
If the benefits were shared across humanity, that could bring us closer to utopia. My worry is that we’ll instead end up with a handful of even wealthier billionaires and millions of people out of work.
I mean that's what all of these execs are openly telling everyone: they want you out of work, they want their ai to be the one to bring the world to it's knees, they want to surveille every second of your day, they want killer drones to use, they want to lay all of your cities to rubble and build "paradises" on top of them like in gaza.
They also openly tell you what they are afraid of btw: collective worker power. something that is massively lacking in our industry, although i feel like it would be one of the easiest industries to unionize in terms of # of workers.
Interesting interview I just watched about how powerful and dangerous these "wishes" or "prophecies" are especially in the hands of the ultra-wealthy: https://www.youtube.com/watch?v=eR7grHa1NR0
> Will these models eventually replace all knowledge work, leaving lawyers, doctors, product managers, software developers, and others out of a job?
Effectively yes, in the current forms. Those professions will likely evolve, but the traditional forms (ie writing code by hand, writing law filings by hand etc) are all dead.
If we have enough energy and raw materials to keep building, yes, it will be an utopia made real. But if there is energy scarcity, then other two outcomes can arise.
Doubt it. When someone can go to Astra MD for 75% of what they used to go to the doctor for, then the remaining doctors only have 25% as many visits. When doctors only have 25% as many visits, they have to compete on price and they make less per visit.
Same argument for plumbers. Everyone always jokes about what a good time it is to be a plumber. But what happens will all the software engineers turn to plumbing? Suddenly it's not such a good time to be a plumber anymore.
What happens when the AI makes a mistake with a diagnosis? I can hear it now 'Oh how perceptive you are, yes that pain in your knee could have been a torn meniscus'.
Does hallucination matter for this application? We've moved beyond raw recall being that important, it seems like for law specifically all relevant facts will be cited and checked easily by humans.
> all relevant facts will be cited and checked easily by humans
I've talked to a lawyer about how they handle this. They do indeed double-check everything, since it'd be embarrassing (or worse) to send hallucinated statements to opposing council or to the court. They still find the assembly a huge time saver
But based on stories in the news on the subject, not everyone has this same level of diligence
Sure but just like generating 100x more code, someone has to review it. So you are wasting everyone in court's time (defendants, prosecutors, judges, staff) by making them parse through what is quite often a bunch of hallucinated slop. Time that could be much better spent on parties who prepared and reviews their own arguments.
The lawyers I know are very fixated on the problem of hallucinated case citations which is amusing to me as a onetime programmer, since case citations have a well-defined syntax and would be relatively easy to check programmatically.
Which makes it all the more bizarre that LLMs have this problem. Claude Code runs the code it generates through a compiler, why can’t an LLM run its product through a cite checker? I’ve seen LLMs fabricate citations.
Interesting to see the callout to companies like harvey in the post itself as consumers rather than competitors? I guess openai isn't quite willing to step into those customer relations themselves?
The product isn’t meant for you or me it is for lawyers. If you can’t take personal liability for a badly written contract then you shouldn’t be using it.
People that are saying OpenAI is screwed because a lack of profit, I'm not of that opinion. They are encroaching on every industry they can. They have name brand recognition, a huge user base and are showing they can be a valuable tool to all types of businesses.
As much as I hate to see it. They are now threatening industries like Engineers, Game Developers, Accountants, 3D modelers, 3D animators, Video Production, Audio Production, Therapist, Tax Auditors, Journalists, Authors, Artists, Mathematicians, Product managers, Every type of analyst and pretty much any other job that can be done behind a computer screen.
Something else to consider is the “experimental” firms they are spinning up through investors to apply their models to industry, eg Shield Technology Partners for IT support. If they can’t sell it to you, they’ll compete against you
Most HNers are clueless that if you have Top Talent + Capital you already have an insurmountable moat. OpenAI, SpaceX, Anthropic all have that and none of the regular guys can compete against them (if they choose to attack that industry)
The legal system, which is a machine/technology by itself, will be eaten out. I wonder what will replace it. Botnet law arbitrage? – Personal assistants constantly negotiating with each other to avoid permanent civil lawsuits?
the cost of making a legal argument can collapse while the cost of reaching an enforceable, legitimate decision may go higher, which will gate the "justice" system even more.
You presented the concern from my adjacent comment perfectly (“LLM performance: common law vs civil law” essentially). So is AI possibly just growing the “Reverence for Professional Experience” factor that plays such a big role for legal compensation here in the US?
It's really not. The legal system is slow and inefficient because it's a deeply pipelined system built to maximize the throughput of the bottleneck resource: judges. Judges are constitutional officers who exercise independent authority in meat space and thus are necessarily limited in number. The rest of the design flows from that.
If you got the judge, all the parties, and all the witnesses in a conference room together until the case was resolved, you could probably handle a lawsuit in a few months. But each judge has hundreds of cases pending before them, so that would never work. Instead, you get something like how a GPU works. You do some work on a case, submit the work to the court, then work on something else for a few months while you wait around to get the results back. Then you do some more work and submit it to the court, then go do something else for a few months while you wait to get the results back. A few months of actual work gets spread out over a few years that way.
Why stop there? The bottleneck is not some immutable force of nature. The number of judges is determined by legislative action. It is within Congress's power to allocate new federal judicial seats, and likewise at the state level with the equivalent lawmaking body (for most states, at least). Why don't they do so?
I find the idea that people can use LLMs to exercise their rights as citizens appealing however. Many people aren't aware of the rights they have, and LLMs are pretty good at surfacing some stuff without having to pay lawyers. Having to hire a lawyer is imo actually a huge way of gatekeeping people from exercising their rights. I heard a lot of local German public institutions are currently being flooded with people arguing their case with the help of LLM that they previously weren't really realistically able to do. So I don't see it all as bad.
>are pretty good at surfacing some stuff without having to pay lawyers
They are also really good at making stuff up as evidenced by the many, many, many examples you read in the news about actual lawyers using AI to write briefs that are full of errors and hallucinations.
In civil courts, you'd likely get more sympathy from a judge if you represented yourself and admitted your lack of understanding, rather than try to appear as someone you're not because you wrote some prompts and copied the output.
Current court systems around the world are just not built to handle the flooding of cases from the citizens.
The stupidest analogy is open source projects having a hard time accepting LLM generated PRs from the masses, because review process is the bottleneck.
No idea how to fix this, to be honest. In coding world, with some mental gymnastics, I can see code not being reviewed by people anymore. In courts, things generally have more consequences, and you can’t really roll back decisions that easily.
I don’t know in the USA but in France, if it’s deemed that you launched a lawsuit knowing very well it wouldn’t succeed, you are susceptible to get a 10k€ fine. Even jail in serious cases.
Yes. And you(r lawyer) can collect lawyer's fees and you can be made to pay the court fees, if you lose.
IME the American legal system is set-up to discourage litigation, though. A common tactic is to bury your opponent in the threat of heavy damages or jail-time to get them to settle for what you were originally after, which courts are perfectly happy to facilitate because it gets a potentially lengthy trial off their dockets. They'll punish (or be biased against) whichever party seems responsible for not accepting a "reasonable" settlement.
Algorithmic abuse of the system to extract payments already exists in the form of the debt collection industry.
Max Junestrand has consistently said Legora treats the model layer as swappable, selecting across frontier providers rather than building the product around one.
OpenAI didn't need to name Legora and Harvey in the second paragraph of the launch post.
They are pre-empting the obvious interpretation of Astra for Law: that moving this far up the legal stack puts them in direct competition with their biggest legal AI customers.
“Don't worry, they can build on us” is a pretty conspicuous message to include on launch day.
They have clearly thought about some pessimistic outcomes.
>Max Junestrand has consistently said Legora treats the model layer as swappable, selecting across frontier providers rather than building the product around one.
Vendor-neutrality for LLMs is such a weak thesis all around, whether for providers or consumers. It weakens the product by being promiscuous and gains no material benefit at all.
LLMs are magic byte(byte) functions, it doesn't make sense to say "we have different providers for magic".
Different flavors of magic require different ritual components and in the name of all that is stable and production worthy, please don't get a necromancer to do your civil construction magic.
Vendor-neutrality helps reduce lock-in, and OpenAI and Anthropic are big enough that the reduction is valuable.
Switching from ChatGPT Enterprise to Legora at my firm was a godsend, it's so much better for legal work, even with the frequent changes to the underlying models.
the thing is a lot of the legal work which will go through this is drafting 100 and 1 variations of draft versions of standard contracts not containing any trade secrets where the contract can be drafted with "replacement/place holder names"
the kind of work mostly done by juniors not yet through their final exam and other "non" lawyers etc.
so it's a slippery slope of "lets just use it for <this> things where it doesn't matter" and then out of laziness and convenience it creeps into all the other places (at least for drafts).
open-ai have proven they can make good / decent models but business strategy is just spray and pray.
they need to pick a lane and optimize for it. coz at their size they can't serve the application layer (a.i startups who can fine-tune models will eat their lunch)
if they gonna do a consumer play - then go ham on that.
otherwise they're gonna get caught in the dreaded middle valley.
Enterprise is eventually get caught (if it isn't already) by the Microsoft/Google. Because with office/teams/suite they were already in every enterprise, and that just added a new tool to existing offerings.
Provisioning and contracts and data retention was just an extension to review of existing ones.
Nobody serious is going to risk sending sensible data to OpenAI/Anthropic, etc because "the benchmarks have shown +8% performance there and +2% there". Irrelevant.
They can't be seen to commit to strongly to a specific product experience, because if they are understood as a regular tech product business that has way different financial scaling considerations than "superintelligent everything-factory"
I suspect that half-assed announcements like this are a result of different people internally with conflicting incentives resulting in a split-the-baby solution.
Wouldn’t this and similar efforts to centralize bureaucracy make AI the new gatekeeper? Without reliable transparent models we’re just trusting OpenAI instead of a hundred top legal firms.
So much for caring about the spirit of the law. Now we'll start an arms race for abusing every possible letter of the law.
It's analogous to crypto. Started from some noble anti-authoritarian ideas and morphed into machine that removes any friction for capital - whoever has the most money will keep gaining the most.
Any lawyers here who have used AI agents heavily for their work? From what I've heard, they're currently very good at searching, analyzing and drafting documents like contracts and patents, but some say they suck at interpreting the law.
They are excellent, especially the latest models. That said, (a) I wouldn't feel safe filing something without a real lawyer looking at it; (b) it can't (easily? legally?) do oral arguments for you; and (c) a lot can happen in the hallways outside the courtroom to move a case forward that the AI can't easily do.
I’m sure there are a wide variety of experiences out there, but here’s my perspective as a former biglaw associate and current solo litigator:
I have had some success using frontier models from the last 6ish months, but only when I can break up my work into discrete and verifiable tasks. For example, I had ~15k pages of discovery I needed to dig through for a summary judgment motion. Instead of just asking Claude to find the best evidence, I asked it first to run a clean, high quality OCR pass (it was almost entirely PDFs). Then I had it generate embeddings and write some reusable python scripts to make keyword and semantic searching easy for agents. While I was writing the brief, I would routinely ask my agent (Claude Code) to use both keyword and semantic searching to find the best evidence supporting whatever assertion I was trying to make. I trusted it because there were traces I could follow.
In other cases/situations, I’ve tried just giving a model access to all the docs and saying “write a brief arguing X,” but it’s always terrible at this. It writes briefs with lots of evocative jargon and rhetorical flourish, but a low signal-to-noise ratio.
Again, I’m sure others’ experiences differ based on workflow, legal area, etc.
Agreed. Six months ago, it was basically a gloried grammarly.
But lately, I’ve been taking hints from the “company brain” models, where it develops a running model of the case, and assesses each new piece as it comes in and updates the file.
I’ve also been using “Ralph Wiggum”-type models where you pass letter or contract drafts back and forth between agents with different goals (rules compliance, grammar, conciseness, ai slop detector, an opposing counsel critic, etc.). After a few rounds, it’s not perfect — but I start with a very good first draft in my hands.
In my experience it basically doesn't even try to interpret the law. It just summarises the publicly available law/guidance out there and, if there is a question about how to interpret some provision, might set out the arguments for each interpretation. It doesn't really take a position. It is pretty good at drafting though. (Legora)
Interesting! In other unrelated domains, models seem more willing to take a position. It may be nuanced, but they do tend to take a stance. I think it's good that it leaves the interpretation to humans, but I wonder if this is also some sort of a guardrail to minimize liability...
I've always said this will be when we get the real Butlerian Jihad, when the AI firms start trying to liquidate the legal profession.
If you automate lawyers out of a job, you can absolutely automate lawmakers out of jobs next. (Not that this would be a bad thing? Maybe pervasive agents for everyone can be the gateway drug to a "this time it's different!" workable direct democracy)
I really hope the bar associations continues to hold lawyers to high standards but I have feeling they may not be ready to handle fallout of AI slop-law.
241 comments
[ 0.22 ms ] story [ 59.1 ms ] threadIf the benefits were shared across humanity, that could bring us closer to utopia. My worry is that we’ll instead end up with a handful of even wealthier billionaires and millions of people out of work.
They also openly tell you what they are afraid of btw: collective worker power. something that is massively lacking in our industry, although i feel like it would be one of the easiest industries to unionize in terms of # of workers.
Interesting interview I just watched about how powerful and dangerous these "wishes" or "prophecies" are especially in the hands of the ultra-wealthy: https://www.youtube.com/watch?v=eR7grHa1NR0
Effectively yes, in the current forms. Those professions will likely evolve, but the traditional forms (ie writing code by hand, writing law filings by hand etc) are all dead.
there will still be writing initial and incremental prompts by hands, until and if LLMs surpass humans in all intellectual functions.
As it stands, it seems far more likely to result in a wonderful life for a few, and an absolute catastrophe for most.
Same argument for plumbers. Everyone always jokes about what a good time it is to be a plumber. But what happens will all the software engineers turn to plumbing? Suddenly it's not such a good time to be a plumber anymore.
https://artificialanalysis.ai/models/gpt-6-astra?omniscience...
Don't be too sure about that. [0]
0: https://www.damiencharlotin.com/hallucinations/
I've talked to a lawyer about how they handle this. They do indeed double-check everything, since it'd be embarrassing (or worse) to send hallucinated statements to opposing council or to the court. They still find the assembly a huge time saver
But based on stories in the news on the subject, not everyone has this same level of diligence
You can get an effectively-zero hallucination rate with the right setup already.
I don't think you can. What I get from this article is that this is not a product they're going to sell to average consumers.
https://commonpaper.com/standards
And yes, even contracts drafted for millions of $ have oversights and unlawful or unenforceable terms.
As much as I hate to see it. They are now threatening industries like Engineers, Game Developers, Accountants, 3D modelers, 3D animators, Video Production, Audio Production, Therapist, Tax Auditors, Journalists, Authors, Artists, Mathematicians, Product managers, Every type of analyst and pretty much any other job that can be done behind a computer screen.
We have big problems for humanity.
[0]https://www.technologyreview.com/2026/06/04/1138391/courts-c...
the cost of making a legal argument can collapse while the cost of reaching an enforceable, legitimate decision may go higher, which will gate the "justice" system even more.
If you got the judge, all the parties, and all the witnesses in a conference room together until the case was resolved, you could probably handle a lawsuit in a few months. But each judge has hundreds of cases pending before them, so that would never work. Instead, you get something like how a GPU works. You do some work on a case, submit the work to the court, then work on something else for a few months while you wait around to get the results back. Then you do some more work and submit it to the court, then go do something else for a few months while you wait to get the results back. A few months of actual work gets spread out over a few years that way.
They are also really good at making stuff up as evidenced by the many, many, many examples you read in the news about actual lawyers using AI to write briefs that are full of errors and hallucinations.
In civil courts, you'd likely get more sympathy from a judge if you represented yourself and admitted your lack of understanding, rather than try to appear as someone you're not because you wrote some prompts and copied the output.
The stupidest analogy is open source projects having a hard time accepting LLM generated PRs from the masses, because review process is the bottleneck.
No idea how to fix this, to be honest. In coding world, with some mental gymnastics, I can see code not being reviewed by people anymore. In courts, things generally have more consequences, and you can’t really roll back decisions that easily.
I don’t know in the USA but in France, if it’s deemed that you launched a lawsuit knowing very well it wouldn’t succeed, you are susceptible to get a 10k€ fine. Even jail in serious cases.
IME the American legal system is set-up to discourage litigation, though. A common tactic is to bury your opponent in the threat of heavy damages or jail-time to get them to settle for what you were originally after, which courts are perfectly happy to facilitate because it gets a potentially lengthy trial off their dockets. They'll punish (or be biased against) whichever party seems responsible for not accepting a "reasonable" settlement.
Algorithmic abuse of the system to extract payments already exists in the form of the debt collection industry.
> API customers including Harvey and Legora will be able to build on Astra for Law, bringing this intelligence into their own products and workflows.
In other words: "no, no, we're not eating our children to prep for the IPO. Don't worry."
OpenAI didn't need to name Legora and Harvey in the second paragraph of the launch post.
They are pre-empting the obvious interpretation of Astra for Law: that moving this far up the legal stack puts them in direct competition with their biggest legal AI customers.
“Don't worry, they can build on us” is a pretty conspicuous message to include on launch day.
They have clearly thought about some pessimistic outcomes.
Vendor-neutrality for LLMs is such a weak thesis all around, whether for providers or consumers. It weakens the product by being promiscuous and gains no material benefit at all.
LLMs are magic byte(byte) functions, it doesn't make sense to say "we have different providers for magic".
Vendor-neutrality helps reduce lock-in, and OpenAI and Anthropic are big enough that the reduction is valuable.
This is everything OpenAI have to say about privacy in this announcement. No guarantees. No promises. Just a pinky-swear promise.
Anyone trusting them–or a lawyer who relies on them–for legal work deserves what they get.
the kind of work mostly done by juniors not yet through their final exam and other "non" lawyers etc.
so it's a slippery slope of "lets just use it for <this> things where it doesn't matter" and then out of laziness and convenience it creeps into all the other places (at least for drafts).
There’s another case making headlines every week.
I get the feeling a lot of them won’t care about this stuff.
they need to pick a lane and optimize for it. coz at their size they can't serve the application layer (a.i startups who can fine-tune models will eat their lunch)
if they gonna do a consumer play - then go ham on that.
otherwise they're gonna get caught in the dreaded middle valley.
Provisioning and contracts and data retention was just an extension to review of existing ones.
Nobody serious is going to risk sending sensible data to OpenAI/Anthropic, etc because "the benchmarks have shown +8% performance there and +2% there". Irrelevant.
I suspect that half-assed announcements like this are a result of different people internally with conflicting incentives resulting in a split-the-baby solution.
People confuse slop with "bad", but slop isn't bad per se, it only becomes bad when real effort was required.
> broadly : a product of little or no value
> food waste (such as garbage) fed to animals
> excreted body waste
https://www.reuters.com/legal/litigation/lawyer-state-farm-f...
So no, I don’t know what they mean.
Is the play here a set of specialized harnesses using their best general model?
It's analogous to crypto. Started from some noble anti-authoritarian ideas and morphed into machine that removes any friction for capital - whoever has the most money will keep gaining the most.
I have had some success using frontier models from the last 6ish months, but only when I can break up my work into discrete and verifiable tasks. For example, I had ~15k pages of discovery I needed to dig through for a summary judgment motion. Instead of just asking Claude to find the best evidence, I asked it first to run a clean, high quality OCR pass (it was almost entirely PDFs). Then I had it generate embeddings and write some reusable python scripts to make keyword and semantic searching easy for agents. While I was writing the brief, I would routinely ask my agent (Claude Code) to use both keyword and semantic searching to find the best evidence supporting whatever assertion I was trying to make. I trusted it because there were traces I could follow.
In other cases/situations, I’ve tried just giving a model access to all the docs and saying “write a brief arguing X,” but it’s always terrible at this. It writes briefs with lots of evocative jargon and rhetorical flourish, but a low signal-to-noise ratio.
Again, I’m sure others’ experiences differ based on workflow, legal area, etc.
But lately, I’ve been taking hints from the “company brain” models, where it develops a running model of the case, and assesses each new piece as it comes in and updates the file.
I’ve also been using “Ralph Wiggum”-type models where you pass letter or contract drafts back and forth between agents with different goals (rules compliance, grammar, conciseness, ai slop detector, an opposing counsel critic, etc.). After a few rounds, it’s not perfect — but I start with a very good first draft in my hands.
Interesting! In other unrelated domains, models seem more willing to take a position. It may be nuanced, but they do tend to take a stance. I think it's good that it leaves the interpretation to humans, but I wonder if this is also some sort of a guardrail to minimize liability...
If you automate lawyers out of a job, you can absolutely automate lawmakers out of jobs next. (Not that this would be a bad thing? Maybe pervasive agents for everyone can be the gateway drug to a "this time it's different!" workable direct democracy)