I think this is probably a good thing. It’s not just public services, but any company that consumers have to deal with (in ways they don’t want to). Historically, only those with the most time, money, connections, and education have been able to extract benefits from these systems. E.g., if someone wants to fight an insurance claim being denied, they might need to hire an expensive lawyer or spend hours performing their own research.
What I’m particularly curious about is what the “endgame” of this looks like. Initially, I imagine we’ll start with LLMs on the consumer side debating LLMs on the service side (this is arguably already the case for a few industries). But at some point, unless identical conditions lead to identical outcomes for the people involved, there will be lawsuits. So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
On the one hand, that has the potential to make things much fairer across economic classes. You don’t get special benefits by having upper class resources. On the other hand, it could lead to increasing the actual requirements for certain benefits, which then reduces class mobility by locking in the benefits to those who already have them, with no clever way to wriggle out of your own level.
I’m just kind of thinking out loud. I don’t know what’s to going to happen long-term, but I think it’s important to monitor these changes carefully as they’re occurring.
> What I’m particularly curious about is what the “endgame” of this looks like.
A higher fraction of money going into the system is lost to navigating adversarial processes. And probably the occasional person getting in legal trouble for not validating their LLM output before feeding it to a formal process.
For customer-funded things (eg home insurance), this means higher prices for the same in-practice benefits (as distinct from nominal benefits; any gap between nominal and actual benefits will probably end up smaller).
For publicly-funded things (government benefits), this could show as increased budgets for the same result or decreased results on the same budget.
.
> So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
Well, maybe standardization of how employers or doctors or whoever are required to report things to the government.
It can also be a way for the public to force the government to perform better, thus lowering the cost of services. For example, the school district in my city is going through a financial crisis of its own making. The public is the only thing holding the district and board of the district accountable, to prevent them from wasting money and creating legal liabilities. AI has been instrumental in allowing the public to better understand the technical financial documents and situations the district is in.
Sure but is that the end game? Systems can respond with things like higher deductible insurance options and graduated assistance based on tax filing which may benefit people who wouldn't file before and also won't use AI for this but were already paying costs of others getting more benefits.
TBH, most government bureaucracy is just pure overhead connected to a jobs program.
A competent government, focused on delivering the services that citizens are actually entitled to, not on paying clerks or avoiding career-ending events at all costs, would do most of this digitally. And by digitally, I don't mean "you can fill the form online, maybe with some auto-fill and autocomplete suggestions for addresses", I mean just sending you the transfer if you're eligible, automatically fetching all the info they need to see if that is the case. If there even is a form, that's too much overhead.
Even Estonia (and other tech-forward EU governments) don't do this. Poland is considered pretty good on the digital front, but many digital processes still result in a (digitally-transmitted) form being printed in some obscure processing center and manually processed by a clerk during business hours.
In my understanding when it comes to doling out benefits, it's not an overhead connected to a jobs program. The less eligible people use benefits, the less the government has to pay out benefits. It's strictly better for the government to make getting benefits as hard as possible.
One could argue that by making benefits hard to access, you under-provision and over-reward. This makes fraud more lucrative getting you into a viscous cycle. Normal people less likely to use benefits and dedicated fraudsters more likely.
I disagree - this doesn’t make it more accessible for people who don’t have the means, it drowns the receivers in bureaucracy and forces them to do orders of magnitude more work out of a lack of understanding. It’s debatable whether or not these systems are built in good faith to allow benefit claimants to appeal or not, but what will happen here is the appeals process will be clamped down on, hard and heavy.
It’s the equivalent of downloading the entire source code of chrome to access a website. are you entitled to do it? Absolutely. Is it a gigantic pain in the ass for the person who has to serve you the 2GB of source code from their git mirror? Yep
Huh, the receivers in this case are the bureaucracy. The senders are the one that has to navigate said bureaucracy.
So, the entire thing is a wash. The systems are complex and confusing, many of them likely by design. Getting an appointment in some of them can take months. Knowing what you actually need can be daunting as quite often there is a byzantine set of requirements that depends on your exact situation and any missed information could take weeks or months to get another visit.
As someone that has worked in a bureaucracy, yes, bureaucrats are themselves victims of bureaucracy. One could even say the primary victims, if you’re going by total exposure time.
Which is why they get paid, of course. I feel more sympathy to the recipients ts who are eligible for benefits but unable to receive them due to onerous paperwork. AI is an equalizer between the desperate amateur and the bureaucratic professionals who know how to effectively navigate their side of the system.
For example, I just hooked an agent up to call my internet company to negotiate a discount. Without that capability, I am either financially taxed by paying too much, or cognitively and time taxed by having to do it myself.
One hypothetical outcome of AI-mediated job loss is that more and more people wind up working in the services (private and public) providing human judgement. Is this person qualified to receive that benefit? Are they truly deserving or just trying to game the system? I wouldn’t be at all surprised if over time we see an explosion in the number of benefits adjudicators or victims services personnel or patient advocates.
Maybe there is an economic equalizer, but I feel like Fable is going to win more appeals than GLM 5.2, so perhaps not truly an equalizer. If the LLM you're using to run your appeal is better than the one your insurance company is using to process your appeal, then you have a better chance of winning. So we have an arms race type situation, and money is always good in arms races.
i agree with most of this, but for appeals and thing where there is a clear correct and fair answer, i think the models converge on the same answer regardless of level of intelligence (e.g., if you ask fable 5 and sonnet 5 what 2+2 is, you’ll get the same output)
That's a good point. It will be interesting to see how it plays out.
In general, I do think this is positive. Companies love putting barriers in front of everything and maybe the AI arms race will result in "fine, you can have a button to cancel" or "fine, we'll approve your medically necessary claim without making you do an appeals dance for our entertainment".
Yeah, this is going to be interesting. First thing I did with the new Instinct bot was have it generate a property tax value appeal for me. Which will almost definitely go through; it's a slam dunk case but I'd been putting it off for an embarrassingly long time. Now it took me 10 minutes on a Sunday. This is obviously good, but I don't know how public agencies are going to deal with the influx of people trying to get the benefits they are owed, now that they can't rely on red tape and busy schedules to decrease the load.
I hope it will force institutions to simplify entitlements. Right now the institutions can pretend like their kafkaesque system is fine because nobody manages to navigate the maze.
> So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
We're seeing how this plays out now in various states that have made their benefits systems nearly impossible to use, which helps them save money by not doling out benefits at all. Having helped some of my elder relatives try to apply for social security payments online makes me very, very concerned about how it's going to look when I'm that age.
The cost of an automated bot that will talk you endlessly in circles until you give up is negligible now.
> What I’m particularly curious about is what the “endgame” of this looks like. Initially, I imagine we’ll start with LLMs on the consumer side debating LLMs on the service side
More data centers, less low-end lawyers, probably more pressure on the judicial system.
I think people using LLMs the right way is great and will democratize information.
However, most of us have experienced people wasting our time, sending us AI slop. People need to learn how to use the tools properly and respect other people's time.
In my view there are three takeaways from this, all of which are correct:
1. Already strained government services need to bear increasing input, affecting waiting times and employee workloads, which is bad.
2. LLMs are empowering citizens to appeal in a social security landscape which often require expert-level knowledge to navigate, which is good.
3.This phenomenon says more about the complexity of our social security systems than it does about LLMs, hopefully prompting our governments to review and start thinking about social security reform.
2b. LLMs are empowering bad actors to exploit loopholes and collect benefits they're not entitled to.
The assumption from a lot of commenters seems to be that these are all legitimate claims and appeals, not a gold rush to get money from the government while the getting's good. In the end it will be the good faith actors that suffer - both those relying on these benefits to survive who will have to jump through more hoops to prove it and those funding the system with their taxes.
Having older parents and watching what my grandparents went thru long before LLMs I just don't worry about the poor old state that much.
The political zeitgeist is to always blame the "bad people", the "welfare queens" and the "undeserving" as the problem with the system. But never the complex system that's seemingly designed to benefit things like corporations extracting as much wealth from the system as they can. Or politicians ear marking the money for things they want to do, rather than the people that put the money in the system.
So, yea, your reasoning just really doesn't hit me in the heartstrings at all.
I decided to read the report this is article is referring to. The report is estimating improper payments which is basically just missing paperwork and other types of payment errors. The report says explicitly that it is not measuring fraud rates.
So I dug a little deeper: roughly 94% of Medicare/Medicaid dollars and 89% of SNAP dollars passed audit reviews in 2025, but I can't find any national percentages for beneficiary fraud rates overall (which is sort of what we are talking about).
Provider fraud does appear to be the main driver of healthcare losses if we look at the enforcement evidence, but that does not mean every improper payment is abuse. It does seem the programs are audited quite extensively (unlike what some other comments in this thread are claiming).
Stressing the benefits system includes improper payments, fraud, and even the perception of fraud or anything else that decreases political support. It also includes the effective denial of service that this paper documents.
It should not be controversial to expect high integrity for a system that pays out money. “Its OK, its just a minimum of $186B in improper payments in 2025, not fraud” shows that it doesn’t have high integrity and that you don’t expect it to.
I think we might be using the term "stress" in different ways.
This paper makes a distinction between direct operational strain and harm created by governments' responses (like adding fees, procedural barriers, and other types of friction in the system). The second part is where my concern is, when bureaucracy is meant to suppress questionable claims it can also delay or deter legitimate benefits.
The paper says it doesn't/can't make definitive claims about AI's impact, but says the strain is moderate. The conclusion does not seem to suggest any severe operational risk to these agencies/systems. Only 9 of 84 potential cases involved benefits or social protection, and it says most cases are good-faith users.
So, I do question what in the paper would show that the burden is large enough to jeopardize benefits? The main finding seems to be that friction in the system may itself effectively deny legitimate access.
Response to your edit: the $186 billion is GAO's government-wide estimate across 64 different programs and not a fraud figure or addressing individual benefits. It contains administrative and document errors, indeterminate payments, and underpayments. And expecting integrity in a system doesn't prevent criticism in the form of asking whether added enforcement bureaucracies cause more harm to legitimate beneficiaries than it prevents in that system.
I don't understand why this is often portrayed negatively. In general it's important that people can defend their rights, and I find it great that LLMs can help a lot with that. One could introduce page/word/character limits, and note that the other side can use LLMs too to summarize/organize/extract. Overall I think it's great that LLMs now help me to understand and rightfully claim things from insurances, companies, etc.
>I don't understand why this is often portrayed negatively. In general it's important that people can defend their rights, and I find it great that LLMs can help a lot with that.
Well I thought it was self evident. It’s portrayed negatively because it causes a denial of service. You should view this positively if you want worse access to public benefits.
Imagine if every time someone reached out to you to ask you a question they did a SAR for all of the information you had on them. They’re technically entitled to do that in many jurisdictions, but it probably won’t help with getting the answer to the question. It’s the same problem as drive by AI PRs except the people receiving them aren’t allowed to ignore them.
This is good I think. I have a friend that worked for a Law Firm / NGO / Charity kind of thing, whose primary purpose was writing appeals for benefits to the VA. These were benefits that the person in question was legally entitled to, but required outside legal assistance to actually attain. Regardless of your personal opinion on defense, VA fraud, etc. it would seem good that people who are legally entitled to a benefit are more easily able to attain the benefit.
LLMs are just revealing how many systems rely on making applications artificially difficult, either to understand or to generate, to reduce volume. These systems have been failing silently for years, unmuting the alarm is the first step to fixing them.
It should be more about how Public Services are underfunded and the process for applying and appeals for benefits is designed to be slow to account for this underfunding, but now something has bypassed one part of the design. It should be an argument as to why we need more funding for public services.
It's becoming obvious that essentially every public-facing organization needs to change. Short story, poetry, essay submissions, comments, reviews, conference paper and journal submissions, polling, job apps, etc.
At this point, I think it's an open question whether there's a scalable way to filter for human effort, that isn't just another layer of AI (e.g. Pangram). I'd love to see something practical come out of this, since the default is probably for public interfaces to just cease to exist, and convert to referral only systems or pure noise.
This is something I have been working on. One of the main issues my users have run into is navigation. The Gov. websites often are not very user friendly; so its hard and time consuming to navigate to and apply for the correct services.
To address this I have users just speak on their "Intent" for example: "I need help with housing and food assistance, my kid has a peanut allergy and I just lost my job so I don't know what to do."
Note: This is a real request
What happens here it a custom program is created, the program has a protocol that shows exactly how the agent navigates the Gov. system and applies for the benefits. Standard modules are applied to actually do the real work of applying, followups etc...
Despite having a lot of professional experience I have not been able to even get a follow-up interview for a job so I am just pouring my efforts into this.
I have been able to help people get housed, apply for benefits, get into shelters and most recently I started working on erpo.law to speed up and reduce the paperwork on all sides for red flag laws.
Lots of the people that have helped me have done so under the table and not at the approval of their agency heads.
Here is a short vid showing a request that locates an existing program, programs can be codified into re-useable units for people w/ common needs or created in real time... Staffed via O*NET roles that hit real providers I minded from other websites.
Note: This one matters to me a lot because my mom was in and out of mental institutions my whole life and I been committed before so I really like helping that population because I get it more than most do.
Im also doing city agencies NYC, Boston, Miami, LA, and Vegas to start because they all have their own tld so its easier for me to filter out the bs traffic and bots and stuff.
In terms of scaling human effort the HITL for each module/protocol-chain step can be delegated where needed and the program will persist that prefrence etc...
Of all the sci-fi things Star Trek predicted, the one that hits hardest is whenever Kirk/Pickard have to have some logical debate with a computer to convince it not to destroy them.
This is the real future, theres no way public services will review every request in the future, everything will be you debating a computer, or you asking your computer to debate their computer.
Treating this as a problem to solve seems like an odd reaction. The public being able to better represent themselves, by making previously difficult content or access paths easier to navigate is a win for the public and therefore a win for the people. Of course, the government may prefer that the public not be able to speak up for itself. This is healthy tension. Honestly, the average person in the public is currently underserved relative to paid lobbyists, resourced activists, lawyers, etc. that know how to engage with the political system in more productive ways.
To me this is one of the reasons that UBI might not be most horrible idea. Sufficiently calibrated UBI would remove this useless work on both sides. Thus making whole system more efficient and maybe even cheaper.
AI used this way is basically anti-compression of a couple bullet points into pages of slop that either have to be parsed by a poor human or worse thrown into another LLM that tries to recover those poorly thought out bullet points from all the noise.
I began receiving public benefits around 2003. Some of those benefits were granted without any contribution from my side at all. Eventually, when we got into the Section 8 housing and other things, there was a lot of paperwork. The application required gathering together a lot of documentation for my ID, my finances, my household makeup. Initially, it was all manual and on paper, and gradually over those two decades, it moved to electronic online processes.
The case workers and bureaucrats who worked on my case always expressed a lot of relief and encouragement that I was filling out the paperwork well. Being that I could do a lot of things electronically, and use my printer effectively, and just that I already had that mindset where I could fill out lots of forms and collate paperwork from a lot of different sources: I had a distinct leg-up compared to other recipients of public benefits.
A lot of people who are poor, illiterate, ESL, and in stressful living situations are not able to keep up with that paperwork and the bureaucratic requirements, like at all, and that is generally a first-order filter for who's not going to receive the entitlements. However, in my experience, there are usually social workers and specialists available, who can assist in the process, and if someone is in actual need of those benefits, not much will stand in their way of obtaining them.
I have also seen a lot of fraud and waste from the provider side of things. However, I do not believe there is a lot of individual fraud. At least not large-scale stuff. Because it would be difficult to carry through with it. If people are writing up appeals with LLMs, they may already be on a razor's edge of denial. It is definitely a real thing that, if you don't need/deserve the benefits, you can legit get buried in paperwork and dropped into bureaucratic mazes of fool's errands and snipe hunts, and you'll never find the end.
So today, the workflow for entitlements and benefits has moved to a completely electronic, online one, and so anyone who can navigate websites and collate PDFs and fill out web forms, these are the people uniquely advantaged to make use of those benefits. It's great that an LLM can write a cogent appeal and a rationale now. I suppose that the government will find themselves summarizing those with LLMs as well. Good luck!
A few weeks ago a guy came up to me at a coffee shop to ask if I wanted to develop his big AI idea that he’s getting patented soon. Obviously I didn’t want to but during the conversation he spent a lot of time telling me that his big thing these days is getting AI systems to find grant opportunities to generate fraudulent applications to. I have a feeling he was lying or delusional, but he said he’s making hundreds of thousands of dollars from this. (I found it amusing in his job recruitment pitch he spent a lot of time bragging about scamming people.)
A complex application process is basically a proof-of-human-work to rate-limit processing costs (for all applications, successful or not) and benefits cost (for successful applications).
In 2026, you can ask an LLM to fill out the application, reducing the expensive proof-of-human-work to a free or cheap machine proof-of-work.
I can immediately think of three different user stories for LLM-generated applications for, without loss of generality, government benefits:
- Alice is honest and likely eligible but didn't apply because the application process was too time-consuming / complex. With LLM assistance, Alice can now complete the process.
- Bob is honest and likely ineligible but didn't apply because he realized he's probably not eligible. With LLM assistance, Bob can now apply anyway -- "Throw spaghetti at the wall and see what sticks."
- Eve is dishonest and ineligible. Eve wants to make money by lying on her application to defraud the government into paying her benefits. With LLM assistance, Eve can more easily make more convincing applications to more agencies.
An interesting question that doesn't seem to be addressed by the paper is how much of the cost is driven by Alice vs. Bob vs. Eve. We can tell very different stories depending on which is the case:
- Mostly Alice: Our current taxes / budget / etc. can only pay for half of what it would cost to pay everyone entitled to benefits under the law. We quietly disenfranchise citizens of their legal right to receive benefits with a bureaucratic process that half of eligible applicants can't complete, but whoops! now a lot more people can complete it with LLM assistance and we're approving most of them because they're legitimately completing the application process and legitimately entitled to benefits. Financially it's killing us to pay what the government promised to all those people to whom it was promised.
- Mostly Bob: A lot more people are applying because LLM's make it super easy for anyone to apply. We don't approve many more applications than we did before -- but each year has about 1,000,000 more applications and 990,000 more denials than the previous year for three years running. Financially it's killing us to pay the processing costs for all those unsuccessful applications.
- Mostly Eve: We're seeing an uptick in applications that are immediately denied, or approved but later revoked, because of fraud. LLM's are making it easier for criminals to defraud the government. Financially it's killing us because we pay those fraudulent benefits before we know they're fraudulent; recovery of funds is slow and uncertain once the fraud is discovered; and fraud imposes additional costs of investigation, prosecution, and punishment.
The people who need this additional help to access water/power/internet/etc often have the most barriers to being able to complete the necessary steps to access benefits. This is a great use of the technology.
This is a bad thing. It means public services will ultimately deploy AI to respond to those requests, and people will be denied instantly with no recourse or human oversight, it will be tyranny of the machine.
"Flooding" implies some nefarious purpose. There's no evidence of that in this paper. Maybe these agents are finally giving people who could never find the time, resources, knowledge, etc. to engage with the system.
"Public services are increasingly becoming more accessible with LLMs" How about that as the title?
They also have zero evidence that this is because of LLMs. They just got some data related to volume, did a really crappy analysis that wouldn't even stand up as an intro to stats course project, and declared their biased opinion as likely true.
"Flooding" could be less nefarious as in the number of requests have risen dramatically in a short period of time quite possibly overwhelming those that are tasked to deal with the requests. Most places have peaks and valleys and can staff appropriately. Having a sudden rise from something unexpected will overwhelm the normal staff. We can debate whether the lack of recognition of the upticks being intentional or not, but possibly the ignoring the uptick is the nefarious bit
I think this is probably a good thing. It’s not just public services, but any company that consumers have to deal with (in ways they don’t want to). Historically, only those with the most time, money, connections, and education have been able to extract benefits from these systems. E.g., if someone wants to fight an insurance claim being denied, they might need to hire an expensive lawyer or spend hours performing their own research.
What I’m particularly curious about is what the “endgame” of this looks like. Initially, I imagine we’ll start with LLMs on the consumer side debating LLMs on the service side (this is arguably already the case for a few industries). But at some point, unless identical conditions lead to identical outcomes for the people involved, there will be lawsuits. So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
On the one hand, that has the potential to make things much fairer across economic classes. You don’t get special benefits by having upper class resources. On the other hand, it could lead to increasing the actual requirements for certain benefits, which then reduces class mobility by locking in the benefits to those who already have them, with no clever way to wriggle out of your own level.
I’m just kind of thinking out loud. I don’t know what’s to going to happen long-term, but I think it’s important to monitor these changes carefully as they’re occurring.
> What I’m particularly curious about is what the “endgame” of this looks like.
A higher fraction of money going into the system is lost to navigating adversarial processes. And probably the occasional person getting in legal trouble for not validating their LLM output before feeding it to a formal process.
For customer-funded things (eg home insurance), this means higher prices for the same in-practice benefits (as distinct from nominal benefits; any gap between nominal and actual benefits will probably end up smaller).
For publicly-funded things (government benefits), this could show as increased budgets for the same result or decreased results on the same budget.
.
> So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
Well, maybe standardization of how employers or doctors or whoever are required to report things to the government.
It can also be a way for the public to force the government to perform better, thus lowering the cost of services. For example, the school district in my city is going through a financial crisis of its own making. The public is the only thing holding the district and board of the district accountable, to prevent them from wasting money and creating legal liabilities. AI has been instrumental in allowing the public to better understand the technical financial documents and situations the district is in.
Sure but is that the end game? Systems can respond with things like higher deductible insurance options and graduated assistance based on tax filing which may benefit people who wouldn't file before and also won't use AI for this but were already paying costs of others getting more benefits.
TBH, most government bureaucracy is just pure overhead connected to a jobs program.
A competent government, focused on delivering the services that citizens are actually entitled to, not on paying clerks or avoiding career-ending events at all costs, would do most of this digitally. And by digitally, I don't mean "you can fill the form online, maybe with some auto-fill and autocomplete suggestions for addresses", I mean just sending you the transfer if you're eligible, automatically fetching all the info they need to see if that is the case. If there even is a form, that's too much overhead.
Even Estonia (and other tech-forward EU governments) don't do this. Poland is considered pretty good on the digital front, but many digital processes still result in a (digitally-transmitted) form being printed in some obscure processing center and manually processed by a clerk during business hours.
In my understanding when it comes to doling out benefits, it's not an overhead connected to a jobs program. The less eligible people use benefits, the less the government has to pay out benefits. It's strictly better for the government to make getting benefits as hard as possible.
One could argue that by making benefits hard to access, you under-provision and over-reward. This makes fraud more lucrative getting you into a viscous cycle. Normal people less likely to use benefits and dedicated fraudsters more likely.
I disagree - this doesn’t make it more accessible for people who don’t have the means, it drowns the receivers in bureaucracy and forces them to do orders of magnitude more work out of a lack of understanding. It’s debatable whether or not these systems are built in good faith to allow benefit claimants to appeal or not, but what will happen here is the appeals process will be clamped down on, hard and heavy.
It’s the equivalent of downloading the entire source code of chrome to access a website. are you entitled to do it? Absolutely. Is it a gigantic pain in the ass for the person who has to serve you the 2GB of source code from their git mirror? Yep
>it drowns the receivers in bureaucracy
Huh, the receivers in this case are the bureaucracy. The senders are the one that has to navigate said bureaucracy.
So, the entire thing is a wash. The systems are complex and confusing, many of them likely by design. Getting an appointment in some of them can take months. Knowing what you actually need can be daunting as quite often there is a byzantine set of requirements that depends on your exact situation and any missed information could take weeks or months to get another visit.
>the receivers in this case are the bureaucracy
As someone that has worked in a bureaucracy, yes, bureaucrats are themselves victims of bureaucracy. One could even say the primary victims, if you’re going by total exposure time.
Which is why they get paid, of course. I feel more sympathy to the recipients ts who are eligible for benefits but unable to receive them due to onerous paperwork. AI is an equalizer between the desperate amateur and the bureaucratic professionals who know how to effectively navigate their side of the system.
For example, I just hooked an agent up to call my internet company to negotiate a discount. Without that capability, I am either financially taxed by paying too much, or cognitively and time taxed by having to do it myself.
The endgame is that everything will be gated by costly signals. What will remain costly as LLM's automate white-collar work?
Responsibility
Status, mostly.
One hypothetical outcome of AI-mediated job loss is that more and more people wind up working in the services (private and public) providing human judgement. Is this person qualified to receive that benefit? Are they truly deserving or just trying to game the system? I wouldn’t be at all surprised if over time we see an explosion in the number of benefits adjudicators or victims services personnel or patient advocates.
I would be astonished to see this outcome. It is costly and highly subjective.
Maybe there is an economic equalizer, but I feel like Fable is going to win more appeals than GLM 5.2, so perhaps not truly an equalizer. If the LLM you're using to run your appeal is better than the one your insurance company is using to process your appeal, then you have a better chance of winning. So we have an arms race type situation, and money is always good in arms races.
i agree with most of this, but for appeals and thing where there is a clear correct and fair answer, i think the models converge on the same answer regardless of level of intelligence (e.g., if you ask fable 5 and sonnet 5 what 2+2 is, you’ll get the same output)
That's a good point. It will be interesting to see how it plays out.
In general, I do think this is positive. Companies love putting barriers in front of everything and maybe the AI arms race will result in "fine, you can have a button to cancel" or "fine, we'll approve your medically necessary claim without making you do an appeals dance for our entertainment".
Yeah, this is going to be interesting. First thing I did with the new Instinct bot was have it generate a property tax value appeal for me. Which will almost definitely go through; it's a slam dunk case but I'd been putting it off for an embarrassingly long time. Now it took me 10 minutes on a Sunday. This is obviously good, but I don't know how public agencies are going to deal with the influx of people trying to get the benefits they are owed, now that they can't rely on red tape and busy schedules to decrease the load.
I hope it will force institutions to simplify entitlements. Right now the institutions can pretend like their kafkaesque system is fine because nobody manages to navigate the maze.
> So I suspect this will lead to a much more formally rigorous system with various “certificates” proving, e.g., that you lost your job or that a doctor has confirmed you have a disability.
We're seeing how this plays out now in various states that have made their benefits systems nearly impossible to use, which helps them save money by not doling out benefits at all. Having helped some of my elder relatives try to apply for social security payments online makes me very, very concerned about how it's going to look when I'm that age.
The cost of an automated bot that will talk you endlessly in circles until you give up is negligible now.
In California I have seen this occur
> various states that have made their benefits systems nearly impossible to use, which helps them save money by not doling out benefits at all
> What I’m particularly curious about is what the “endgame” of this looks like. Initially, I imagine we’ll start with LLMs on the consumer side debating LLMs on the service side
More data centers, less low-end lawyers, probably more pressure on the judicial system.
I think people using LLMs the right way is great and will democratize information.
However, most of us have experienced people wasting our time, sending us AI slop. People need to learn how to use the tools properly and respect other people's time.
Now read the part where Voter Roll Purge request portal was flooded.
In my view there are three takeaways from this, all of which are correct:
1. Already strained government services need to bear increasing input, affecting waiting times and employee workloads, which is bad.
2. LLMs are empowering citizens to appeal in a social security landscape which often require expert-level knowledge to navigate, which is good.
3.This phenomenon says more about the complexity of our social security systems than it does about LLMs, hopefully prompting our governments to review and start thinking about social security reform.
>2. LLMs are empowering citizens to appeal in a social security landscape which often require expert-level knowledge to navigate, which is good.
Including Voter Roll Purge requests? That’s in the paper as well.
Child protective services has a tip line.
You forgot:
2b. LLMs are empowering bad actors to exploit loopholes and collect benefits they're not entitled to.
The assumption from a lot of commenters seems to be that these are all legitimate claims and appeals, not a gold rush to get money from the government while the getting's good. In the end it will be the good faith actors that suffer - both those relying on these benefits to survive who will have to jump through more hoops to prove it and those funding the system with their taxes.
Having older parents and watching what my grandparents went thru long before LLMs I just don't worry about the poor old state that much.
The political zeitgeist is to always blame the "bad people", the "welfare queens" and the "undeserving" as the problem with the system. But never the complex system that's seemingly designed to benefit things like corporations extracting as much wealth from the system as they can. Or politicians ear marking the money for things they want to do, rather than the people that put the money in the system.
So, yea, your reasoning just really doesn't hit me in the heartstrings at all.
>Having older parents and watching what my grandparents went thru long before LLMs I just don't worry about the poor old state that much.
You should worry more about your parents whose benefits are jeopardized by stress to the benefits system.
Is there any evidence that this is actually a real worry?
Don’t audits usually find that vast majority of benefits are valid and correctly applied?
Medicaid and medicare make up 50% of the "abuse" (primarily overpayment), which again points towards companies milking the system, not individuals https://washingtonstand.com/article/gao-finds-15-major-feder...
I decided to read the report this is article is referring to. The report is estimating improper payments which is basically just missing paperwork and other types of payment errors. The report says explicitly that it is not measuring fraud rates.
So I dug a little deeper: roughly 94% of Medicare/Medicaid dollars and 89% of SNAP dollars passed audit reviews in 2025, but I can't find any national percentages for beneficiary fraud rates overall (which is sort of what we are talking about).
Provider fraud does appear to be the main driver of healthcare losses if we look at the enforcement evidence, but that does not mean every improper payment is abuse. It does seem the programs are audited quite extensively (unlike what some other comments in this thread are claiming).
Stress to the benefits system != fraud
Stressing the benefits system includes improper payments, fraud, and even the perception of fraud or anything else that decreases political support. It also includes the effective denial of service that this paper documents.
It should not be controversial to expect high integrity for a system that pays out money. “Its OK, its just a minimum of $186B in improper payments in 2025, not fraud” shows that it doesn’t have high integrity and that you don’t expect it to.
I think we might be using the term "stress" in different ways.
This paper makes a distinction between direct operational strain and harm created by governments' responses (like adding fees, procedural barriers, and other types of friction in the system). The second part is where my concern is, when bureaucracy is meant to suppress questionable claims it can also delay or deter legitimate benefits.
The paper says it doesn't/can't make definitive claims about AI's impact, but says the strain is moderate. The conclusion does not seem to suggest any severe operational risk to these agencies/systems. Only 9 of 84 potential cases involved benefits or social protection, and it says most cases are good-faith users.
So, I do question what in the paper would show that the burden is large enough to jeopardize benefits? The main finding seems to be that friction in the system may itself effectively deny legitimate access.
Response to your edit: the $186 billion is GAO's government-wide estimate across 64 different programs and not a fraud figure or addressing individual benefits. It contains administrative and document errors, indeterminate payments, and underpayments. And expecting integrity in a system doesn't prevent criticism in the form of asking whether added enforcement bureaucracies cause more harm to legitimate beneficiaries than it prevents in that system.
No one ever audits them.
That can’t possibly be true since I’ve read audit reports on public benefits
>But never the complex system that's seemingly designed to benefit things like corporations extracting as much wealth from the system as they can.
What “system”? Are you talking about? Social security? How is social security designed to maximize corporate wealth extraction?
If you mean the more nebulous “system” (“the system, man”), or capitalism, etc. then you’re equivocating.
is that actually happenng?
"No, but we're reporting it is" --southpark
> hopefully prompting our governments to review and start thinking about social security reform
I see the opposite, where LLMs and purchased datasets will make it easier for governments to create additional barriers for taxpayers seeking welfare.
I don't understand why this is often portrayed negatively. In general it's important that people can defend their rights, and I find it great that LLMs can help a lot with that. One could introduce page/word/character limits, and note that the other side can use LLMs too to summarize/organize/extract. Overall I think it's great that LLMs now help me to understand and rightfully claim things from insurances, companies, etc.
To be fair, the source mentions the upside ("Although improving Service accessibility is beneficial [...]")
Seems like bureaucracy was added as a subtle way to introduce rate limits, and AI is challenging that mechanism.
> Seems like bureaucracy was added as NOT a subtle way to introduce rate limits
some teams get incentives when the total awards go down in a stable, badly run bureaucracy.. this is real
LLMs are a magic machine to turn commons into tragedies.
Great quote.
Good luck trying to get your book published.
Or run an open source repo from community contributions.
>I don't understand why this is often portrayed negatively. In general it's important that people can defend their rights, and I find it great that LLMs can help a lot with that.
Well I thought it was self evident. It’s portrayed negatively because it causes a denial of service. You should view this positively if you want worse access to public benefits.
Imagine if every time someone reached out to you to ask you a question they did a SAR for all of the information you had on them. They’re technically entitled to do that in many jurisdictions, but it probably won’t help with getting the answer to the question. It’s the same problem as drive by AI PRs except the people receiving them aren’t allowed to ignore them.
This is good I think. I have a friend that worked for a Law Firm / NGO / Charity kind of thing, whose primary purpose was writing appeals for benefits to the VA. These were benefits that the person in question was legally entitled to, but required outside legal assistance to actually attain. Regardless of your personal opinion on defense, VA fraud, etc. it would seem good that people who are legally entitled to a benefit are more easily able to attain the benefit.
LLMs are just revealing how many systems rely on making applications artificially difficult, either to understand or to generate, to reduce volume. These systems have been failing silently for years, unmuting the alarm is the first step to fixing them.
It should be more about how Public Services are underfunded and the process for applying and appeals for benefits is designed to be slow to account for this underfunding, but now something has bypassed one part of the design. It should be an argument as to why we need more funding for public services.
Or, "LLMs making it increasingly hard for government to categorically deny access to public services" depending on how you want to present the issue.
It's becoming obvious that essentially every public-facing organization needs to change. Short story, poetry, essay submissions, comments, reviews, conference paper and journal submissions, polling, job apps, etc.
At this point, I think it's an open question whether there's a scalable way to filter for human effort, that isn't just another layer of AI (e.g. Pangram). I'd love to see something practical come out of this, since the default is probably for public interfaces to just cease to exist, and convert to referral only systems or pure noise.
This is something I have been working on. One of the main issues my users have run into is navigation. The Gov. websites often are not very user friendly; so its hard and time consuming to navigate to and apply for the correct services. To address this I have users just speak on their "Intent" for example: "I need help with housing and food assistance, my kid has a peanut allergy and I just lost my job so I don't know what to do."
Note: This is a real request
What happens here it a custom program is created, the program has a protocol that shows exactly how the agent navigates the Gov. system and applies for the benefits. Standard modules are applied to actually do the real work of applying, followups etc...
Despite having a lot of professional experience I have not been able to even get a follow-up interview for a job so I am just pouring my efforts into this.
I have been able to help people get housed, apply for benefits, get into shelters and most recently I started working on erpo.law to speed up and reduce the paperwork on all sides for red flag laws.
Lots of the people that have helped me have done so under the table and not at the approval of their agency heads.
Here is a short vid showing a request that locates an existing program, programs can be codified into re-useable units for people w/ common needs or created in real time... Staffed via O*NET roles that hit real providers I minded from other websites.
Video: https://youtu.be/JOjG71i2F5E
Note: This one matters to me a lot because my mom was in and out of mental institutions my whole life and I been committed before so I really like helping that population because I get it more than most do.
Im also doing city agencies NYC, Boston, Miami, LA, and Vegas to start because they all have their own tld so its easier for me to filter out the bs traffic and bots and stuff.
In terms of scaling human effort the HITL for each module/protocol-chain step can be delegated where needed and the program will persist that prefrence etc...
Seems plainly obvious to me as well
Of all the sci-fi things Star Trek predicted, the one that hits hardest is whenever Kirk/Pickard have to have some logical debate with a computer to convince it not to destroy them.
This is the real future, theres no way public services will review every request in the future, everything will be you debating a computer, or you asking your computer to debate their computer.
Treating this as a problem to solve seems like an odd reaction. The public being able to better represent themselves, by making previously difficult content or access paths easier to navigate is a win for the public and therefore a win for the people. Of course, the government may prefer that the public not be able to speak up for itself. This is healthy tension. Honestly, the average person in the public is currently underserved relative to paid lobbyists, resourced activists, lawyers, etc. that know how to engage with the political system in more productive ways.
To me this is one of the reasons that UBI might not be most horrible idea. Sufficiently calibrated UBI would remove this useless work on both sides. Thus making whole system more efficient and maybe even cheaper.
AI used this way is basically anti-compression of a couple bullet points into pages of slop that either have to be parsed by a poor human or worse thrown into another LLM that tries to recover those poorly thought out bullet points from all the noise.
I began receiving public benefits around 2003. Some of those benefits were granted without any contribution from my side at all. Eventually, when we got into the Section 8 housing and other things, there was a lot of paperwork. The application required gathering together a lot of documentation for my ID, my finances, my household makeup. Initially, it was all manual and on paper, and gradually over those two decades, it moved to electronic online processes.
The case workers and bureaucrats who worked on my case always expressed a lot of relief and encouragement that I was filling out the paperwork well. Being that I could do a lot of things electronically, and use my printer effectively, and just that I already had that mindset where I could fill out lots of forms and collate paperwork from a lot of different sources: I had a distinct leg-up compared to other recipients of public benefits.
A lot of people who are poor, illiterate, ESL, and in stressful living situations are not able to keep up with that paperwork and the bureaucratic requirements, like at all, and that is generally a first-order filter for who's not going to receive the entitlements. However, in my experience, there are usually social workers and specialists available, who can assist in the process, and if someone is in actual need of those benefits, not much will stand in their way of obtaining them.
I have also seen a lot of fraud and waste from the provider side of things. However, I do not believe there is a lot of individual fraud. At least not large-scale stuff. Because it would be difficult to carry through with it. If people are writing up appeals with LLMs, they may already be on a razor's edge of denial. It is definitely a real thing that, if you don't need/deserve the benefits, you can legit get buried in paperwork and dropped into bureaucratic mazes of fool's errands and snipe hunts, and you'll never find the end.
So today, the workflow for entitlements and benefits has moved to a completely electronic, online one, and so anyone who can navigate websites and collate PDFs and fill out web forms, these are the people uniquely advantaged to make use of those benefits. It's great that an LLM can write a cogent appeal and a rationale now. I suppose that the government will find themselves summarizing those with LLMs as well. Good luck!
A few weeks ago a guy came up to me at a coffee shop to ask if I wanted to develop his big AI idea that he’s getting patented soon. Obviously I didn’t want to but during the conversation he spent a lot of time telling me that his big thing these days is getting AI systems to find grant opportunities to generate fraudulent applications to. I have a feeling he was lying or delusional, but he said he’s making hundreds of thousands of dollars from this. (I found it amusing in his job recruitment pitch he spent a lot of time bragging about scamming people.)
A complex application process is basically a proof-of-human-work to rate-limit processing costs (for all applications, successful or not) and benefits cost (for successful applications).
In 2026, you can ask an LLM to fill out the application, reducing the expensive proof-of-human-work to a free or cheap machine proof-of-work.
I can immediately think of three different user stories for LLM-generated applications for, without loss of generality, government benefits:
- Alice is honest and likely eligible but didn't apply because the application process was too time-consuming / complex. With LLM assistance, Alice can now complete the process.
- Bob is honest and likely ineligible but didn't apply because he realized he's probably not eligible. With LLM assistance, Bob can now apply anyway -- "Throw spaghetti at the wall and see what sticks."
- Eve is dishonest and ineligible. Eve wants to make money by lying on her application to defraud the government into paying her benefits. With LLM assistance, Eve can more easily make more convincing applications to more agencies.
An interesting question that doesn't seem to be addressed by the paper is how much of the cost is driven by Alice vs. Bob vs. Eve. We can tell very different stories depending on which is the case:
- Mostly Alice: Our current taxes / budget / etc. can only pay for half of what it would cost to pay everyone entitled to benefits under the law. We quietly disenfranchise citizens of their legal right to receive benefits with a bureaucratic process that half of eligible applicants can't complete, but whoops! now a lot more people can complete it with LLM assistance and we're approving most of them because they're legitimately completing the application process and legitimately entitled to benefits. Financially it's killing us to pay what the government promised to all those people to whom it was promised.
- Mostly Bob: A lot more people are applying because LLM's make it super easy for anyone to apply. We don't approve many more applications than we did before -- but each year has about 1,000,000 more applications and 990,000 more denials than the previous year for three years running. Financially it's killing us to pay the processing costs for all those unsuccessful applications.
- Mostly Eve: We're seeing an uptick in applications that are immediately denied, or approved but later revoked, because of fraud. LLM's are making it easier for criminals to defraud the government. Financially it's killing us because we pay those fraudulent benefits before we know they're fraudulent; recovery of funds is slow and uncertain once the fraud is discovered; and fraud imposes additional costs of investigation, prosecution, and punishment.
When an immovable object meets an unstoppable force.
Hackernews wrestles with AI slop vs. public benefits.
The people who need this additional help to access water/power/internet/etc often have the most barriers to being able to complete the necessary steps to access benefits. This is a great use of the technology.
This is a bad thing. It means public services will ultimately deploy AI to respond to those requests, and people will be denied instantly with no recourse or human oversight, it will be tyranny of the machine.
Was it intentionally made difficult to communicate with the government and LLMs are just not sensitive to the artificial friction?
"Flooding" implies some nefarious purpose. There's no evidence of that in this paper. Maybe these agents are finally giving people who could never find the time, resources, knowledge, etc. to engage with the system.
"Public services are increasingly becoming more accessible with LLMs" How about that as the title?
They also have zero evidence that this is because of LLMs. They just got some data related to volume, did a really crappy analysis that wouldn't even stand up as an intro to stats course project, and declared their biased opinion as likely true.
"Flooding" could be less nefarious as in the number of requests have risen dramatically in a short period of time quite possibly overwhelming those that are tasked to deal with the requests. Most places have peaks and valleys and can staff appropriately. Having a sudden rise from something unexpected will overwhelm the normal staff. We can debate whether the lack of recognition of the upticks being intentional or not, but possibly the ignoring the uptick is the nefarious bit
AI destroys everything. It's such spammy trash.
>Reduce the Expected Benefit.
This could be the chance to reverse America out of becoming a welfare state.
Stop corporate welfare first.