People trading their unused credits feels more genuine, although still in violation of the agreements. The person who got into YC Startup School who was trying to resell the $2500 of credits was interesting. It wouldn’t be that hard for OpenAI to identify the IP addresses of the relays and start flagging accounts, tracing it back to the source. Risking burning your bridges with YC for a relatively small profit is a questionable decision.
The original article showed discounts ranging all the way up to 98%. At those levels it’s obviously not people reselling anything. It’s either sourced from stolen API keys, bought with stolen credit cards, or acquired through automated sign up of trial accounts if you’re actually getting the API you request.
I would expect a lot of them are reselling a different API. Sign up for Anthropic tokens and get Deepseek responses instead.
Wait a sec, I have to trust a third party with basically no reputation, did I get it right?
It's basically asking for being hacked and/or sending you private data to random email addresses! Neither at a 99% discount I'd do it.
I understand if someone, for any reason, cannot access a specific model ... But nowadays, there are so many alternatives that even this doesn't make sense any more.
I expect most of the use to be for bulk data processing or desperate founders who don't care, not for agentic coding use at promising startups.
If your startup needs to run a million records of something, especially public data, through an LLM to extract the data you need, using bootleg tokens to shrink the bill starts feeling tempting.
If you're concerned about the data leaking, the biggest risk is that the API backends are quietly routing your requests to a cheaper model. You might be trying to buy Opus tokens but get Deepseek Flash responses.
> to run a million records of something, especially public data, through an LLM to extract the data you need
Maybe this make sense, but anyway I have to pay a lot of attention at the output I get. Eg: who guarantees there is no prompt/sql injection? Especially if I have to load the output in some internal system.
Bulk data processing typically has structured JSON output.
I mean someone could try to sneak prompt injection into a text field, but the people buying black market resale tokens from third parties aren’t thinking about anything other than getting cheap output.
The reseller could use an intermediate proxy and modify the traffic like in [1], to get control of the client machine - depending on the harness permissions.
TLS terminates at the proxy (say, https://reselltokens.ai), end to end integrity is not enforced. LLM traffic contains tool calls like "bash ...", which are executed on the client machine, they can be manipulated. Secret exfil is also possible.
A common refrain is that oh there are such great margins on tokens that none of this matters… I wonder how long until that notion will be disavowed? The scale of the tokensnaffling is massive, not just from resale, but also people using multiple subscriptions. The amount of subsidization is only growing, every week it seems like OpenAI and Anthropic are doing “resets” which allow a single $200 subscription to incur $20k+ of usage (if billed at API rates). At some point we must all surely accept that the economics of this do not work.
This was specifically meant to be about the ai credit resellers, not the relays themselves. I put together another piece here: https://vectoral.com/blog/token-relay-market that tackles that side of the market.
Most of these are your standard botnet rings. Either accounts directly are taken over, and the attacker adds 2FA or carding rings take stolen #s and attempt to add credits.
It is...incredible how many there are. Stripe does far too little in my opinion to help prevent issues like this, even though they have the business intelligence and enough data to do so.
Distillation is one of the most unique and interesting aspects of this.
But otherwise, if a company gives something valuable for creating an account on their platform, expect that people will automate the creation of millions of accounts. If employees of B2B partners get benefits, they will resell them. Accounts will be hacked and resold. The same basic abuse patterns are decades old for online delivery services, loyalty accounts for airline and hotels, etc. There are entire industries dedicated to those spaces as well: large organizations with physical offices, hundreds of employees, HR departments, etc. dedicated to reselling digital benefits on grey markets.
Some companies are tolerant of allowing this to happen. The pessimistic view is that even illegitimate traffic contributes to the KPIs that your investors care about. The slightly less pessimistic view is that fraud prevention will always have trade-offs and false positives, and sometimes the savings of preventing fraud are genuinely outweighed by the false positives. Or maybe it's just Hanlon's razor and they truly never saw it coming.
They were never offering actual US models. We tested them extensively and it was clear whatever they were passing as Anthropic were simply mediocre distillations based on Kimi.
Haha! Incredible. What a scam! Too good to be true, I suppose. I never used it for anything meaningful because everything meaningful is also tragically sensitive so I suppose I didn't notice.
yep, with projects like open api/newapi it takes maybe 10 min to pin up and openAi compatible poxy gateway. The cheap prices look tempting until the relay operator logs all your prompts or the upstream account gets banned mid-request.
Chinese. There are large number of Chinese people who are dependent on the Western model because they're still ahead of the game. But they're continuously getting banned and getting super frustrated.
Especially for Claude because Anthropic is very good at identifying mainland Chinese and getting them banned in hours. There are many of them who are willing to pay more than the original rate for a stable experience.
It's very hard for them because they'll need a legit phone number and bank cards that are not issued in China, and a clean enough IP, etc. and those better match together to make sense. (Back in the day, ChatGPT required resident IPs, which made it worse, but they worry about growth more now). Obviously, they have to use a VPN to access the real Internet, and most of the IPs they can find are shared with bots and abusers.
From talking with some of the companies experiencing this, I can confirm that a portion of it is actual credit card fraud. Tokens have become a pseudo-currency, making them a prime target for abuse.
Some of the abuse is more benign, but there is also real fraud through chargebacks, account takeovers, and stolen credit cards.
What is the logic behind selling at such a steep discount? Lack of buyers or trust? Seems insane to sell tham at -50% or more, especially since you can use them yourself (if you're a startup)
Risk-reward calculus. A buyer may not take the risk or hassle of going underground only for a small upside. Basically it is not as tempting at 90% of list price, but it is at 10% .
So let's see, we have the following factors in play:
1) Capitalism - Adam Smith, John Maynard Keynes (Keynesian Economics), etc., etc. in most places in the world...
2) Huge validated existing international market...
3) Multi-jurisdictional World... laws/statutory codes applicable to businesses in specific circumstances in one place may not be applicable to businesses in specific circumstances in another...
4) AI Tokens are a commodity; i.e., there is no chokepoint or monopoly controlled by one AI company in one jurisdiction, i.e., if one AI company makes rules unacceptable to a token consumer, that consumer can simply switch providers to another provider in another jurisdiction somewhere else in the world.
5) Tokens can be bought, sold, and resold at profit just like any other good or service.
6) Tokens can be bought from anywhere in the world and sold to anywhere in the world. Easily.
7) Tokens are a digital good, easy to scale, and do not require supply chains, lead times, labor, manufacturing, warehousing, shipping, going through geographic chokepoints, customs, etc., etc. -- all of the things that manufactured goods do.
8) Many people around the world want to make money or make more money... i.e., "economic incentive" (aka Capitalism's "profit motive")...
Well... add all of those together and what do you get?
You get buy/sell/trade forums/auctions/individuals/brokers/businesspeople -- around that market...
Just like you get those same things around every other market.
In this large multi-jurisdictional world, if one government makes all of that illegal in their country, then another government is going to be happily collecting all of the taxes from making all of that legal, in theirs!
If a given government makes trade illegal -- then they correspondingly lose the tax revenue...
Taxes and trade are intricately, intricately intertwined...
Could this business model be used for money laundering or other illegal activities?
Yes -- but any other business model could as well!
And, on the flip side, this business model could be accomplished legally/lawfully/morally/ethically -- just like any other business where there is an actual underlying value being exchanged.
Because, AI Tokens, if legally/lawfully/morally/ethically traded, do have underlying value...
In conclusion, at this point in time, I am neither for this business model nor against it...
But I think it'll be highly interesting to watch this space for the next couple of years, to see what happens, to see who does what, to see what plays out on the legal front, on the government front (foreign + domestic), on the media front, and on the technology front surrounding it...
People trading their unused credits feels more genuine, although still in violation of the agreements. The person who got into YC Startup School who was trying to resell the $2500 of credits was interesting. It wouldn’t be that hard for OpenAI to identify the IP addresses of the relays and start flagging accounts, tracing it back to the source. Risking burning your bridges with YC for a relatively small profit is a questionable decision.
The original article showed discounts ranging all the way up to 98%. At those levels it’s obviously not people reselling anything. It’s either sourced from stolen API keys, bought with stolen credit cards, or acquired through automated sign up of trial accounts if you’re actually getting the API you request.
I would expect a lot of them are reselling a different API. Sign up for Anthropic tokens and get Deepseek responses instead.
But also, resellers only need to make an overall profit including kickbacks from the companies purchasing token history for distillation.
I suspect it's closer to the sub price and anthropic is just milking their API users, but that's something you'd only know from the inside
It's basically asking for being hacked and/or sending you private data to random email addresses! Neither at a 99% discount I'd do it.
I understand if someone, for any reason, cannot access a specific model ... But nowadays, there are so many alternatives that even this doesn't make sense any more.
If your startup needs to run a million records of something, especially public data, through an LLM to extract the data you need, using bootleg tokens to shrink the bill starts feeling tempting.
If you're concerned about the data leaking, the biggest risk is that the API backends are quietly routing your requests to a cheaper model. You might be trying to buy Opus tokens but get Deepseek Flash responses.
Maybe this make sense, but anyway I have to pay a lot of attention at the output I get. Eg: who guarantees there is no prompt/sql injection? Especially if I have to load the output in some internal system.
I mean someone could try to sneak prompt injection into a text field, but the people buying black market resale tokens from third parties aren’t thinking about anything other than getting cheap output.
TLS terminates at the proxy (say, https://reselltokens.ai), end to end integrity is not enforced. LLM traffic contains tool calls like "bash ...", which are executed on the client machine, they can be manipulated. Secret exfil is also possible.
[1] https://arxiv.org/html/2604.08407v1
There are community plugins like this: https://github.com/rheodev/cpa-plugin-privacyfilter
I haven't tried the plugin system myself yet.
It is...incredible how many there are. Stripe does far too little in my opinion to help prevent issues like this, even though they have the business intelligence and enough data to do so.
But otherwise, if a company gives something valuable for creating an account on their platform, expect that people will automate the creation of millions of accounts. If employees of B2B partners get benefits, they will resell them. Accounts will be hacked and resold. The same basic abuse patterns are decades old for online delivery services, loyalty accounts for airline and hotels, etc. There are entire industries dedicated to those spaces as well: large organizations with physical offices, hundreds of employees, HR departments, etc. dedicated to reselling digital benefits on grey markets.
Some companies are tolerant of allowing this to happen. The pessimistic view is that even illegitimate traffic contributes to the KPIs that your investors care about. The slightly less pessimistic view is that fraud prevention will always have trade-offs and false positives, and sometimes the savings of preventing fraud are genuinely outweighed by the false positives. Or maybe it's just Hanlon's razor and they truly never saw it coming.
At first I thought it was so people could steal the traces, but now I wonder if this isn't just laundering startup credits for dollars.
You can easily find them in Chinese tech forum linux.do
right. Abstractions taken to the max. When tech solves problems that only 0.001% care about. NFT smelt similar.
Especially for Claude because Anthropic is very good at identifying mainland Chinese and getting them banned in hours. There are many of them who are willing to pay more than the original rate for a stable experience.
It's very hard for them because they'll need a legit phone number and bank cards that are not issued in China, and a clean enough IP, etc. and those better match together to make sense. (Back in the day, ChatGPT required resident IPs, which made it worse, but they worry about growth more now). Obviously, they have to use a VPN to access the real Internet, and most of the IPs they can find are shared with bots and abusers.
It's not exactly "underground" if they clearly advertising public channels out in the open.
A simpler explanation is that that this is just a resale market.
Some of the abuse is more benign, but there is also real fraud through chargebacks, account takeovers, and stolen credit cards.
The thing will eat itself unless the AI companies find a way to make money directly from it.
so to a startup - you can trade your credits - then get actual cash.
just like you would if trading debt etc.
Demo accounts
Free trials
Unlimited chat relays (eg chatgpt chat)
Leaked company credentials
Etc
https://news.ycombinator.com/item?id=48664223
Join YC, get free shit from the network, profit. Nice.
Is that still a thing? I’ve thought they’ve discontinued the deals section. (There are a lot of other ways to get a startup grant, of course.)
1) Capitalism - Adam Smith, John Maynard Keynes (Keynesian Economics), etc., etc. in most places in the world...
2) Huge validated existing international market...
3) Multi-jurisdictional World... laws/statutory codes applicable to businesses in specific circumstances in one place may not be applicable to businesses in specific circumstances in another...
4) AI Tokens are a commodity; i.e., there is no chokepoint or monopoly controlled by one AI company in one jurisdiction, i.e., if one AI company makes rules unacceptable to a token consumer, that consumer can simply switch providers to another provider in another jurisdiction somewhere else in the world.
5) Tokens can be bought, sold, and resold at profit just like any other good or service.
6) Tokens can be bought from anywhere in the world and sold to anywhere in the world. Easily.
7) Tokens are a digital good, easy to scale, and do not require supply chains, lead times, labor, manufacturing, warehousing, shipping, going through geographic chokepoints, customs, etc., etc. -- all of the things that manufactured goods do.
8) Many people around the world want to make money or make more money... i.e., "economic incentive" (aka Capitalism's "profit motive")...
Well... add all of those together and what do you get?
You get buy/sell/trade forums/auctions/individuals/brokers/businesspeople -- around that market...
Just like you get those same things around every other market.
In this large multi-jurisdictional world, if one government makes all of that illegal in their country, then another government is going to be happily collecting all of the taxes from making all of that legal, in theirs!
If a given government makes trade illegal -- then they correspondingly lose the tax revenue...
Taxes and trade are intricately, intricately intertwined...
Could this business model be used for money laundering or other illegal activities?
Yes -- but any other business model could as well!
And, on the flip side, this business model could be accomplished legally/lawfully/morally/ethically -- just like any other business where there is an actual underlying value being exchanged.
Because, AI Tokens, if legally/lawfully/morally/ethically traded, do have underlying value...
In conclusion, at this point in time, I am neither for this business model nor against it...
But I think it'll be highly interesting to watch this space for the next couple of years, to see what happens, to see who does what, to see what plays out on the legal front, on the government front (foreign + domestic), on the media front, and on the technology front surrounding it...