Tell HN: OpenAI now requires ID verification and won't refund API credits
87 comments
·October 25, 2025deaux
When they started doing this ID verification early this year, I expressed outrage on here, and was met by comments downplaying it saying "It's a given that soon the others will follow". I'm sure some of those came from people at OpenAI.
We're now at the end of the year and neither Google nor Anthropic nor any single other LLM provider does this. OpenAI does this because their CEO is SamA. That's it.
rsync
A certain business I own has an openai account for testing and research purposes.
What ID would we provide?
Would we pick some random employee to attach to the account?
What relevance does this have to the notion of “piercing the corporate veil” if a business account is tied to someone’s drivers license?
I place the blame for this situation squarely on the careless and thoughtless user population who have blindly provided their phone numbers and now ID scans to any old random, fly by night, start up who request them.
paulddraper
I assume the correct answer is an officer of the company, the same as for who signs contracts etc
syntaxing
You can try the latest GLM 4.6 https://z.ai/ . Their coding plan is $6 a month and performs on par to Sonnet 4 for my personal task. Sonnet 4.5 still has an edge though. All of ZLM’s models are also open sourced so you can run it locally if you want
mark_l_watson
I am mostly retired but I am thinking of restarting a solo products mini-company next year. I have been looking at much less expensive options like Alibaba Cloud, GLM, Kimi K2, etc. There is a recent Stanford study showing most US startups are using less expensive Chinese models, but I think usually hosted in the US.
For now I am happy enough with Gemini and GPT-5 because my usage is so lite that anything is cheap. For many engineering use cases, Gemini-2.5-flash-lite works well enough.
How do you use GLM? With codex —oss? Or, just ‘raw’ with no agent-wrapping coding environment?
syntaxing
I use it directly with Claude code [1]. Honestly, it just makes sense IMO to host your own model when you have your own company. You can try something like openrouter for now and then setup your own hardware. Since most of these models are MoE, you dont have to load everything in VRAM. A mixture of a 5090 + EPYC CPU + 256GB of DDR5 RAM can go a very long way. You can unload most of the expert layers onto CPU and leave the rest on GPU. As usual Unsloth has a great page about it [2]
[1] https://docs.z.ai/scenario-example/develop-tools/claude [2] https://docs.unsloth.ai/models/glm-4.6-how-to-run-locally
mitjam
Hope you‘ll share your story if you start. Love your book on langchain from iirc 2y ago, it got me going.
mistrial9
> There is a recent Stanford study showing most US startups are using less expensive Chinese models
link ?
mitjam
Idk if this is the reference but it’s in the same direction:
„ These days, when entrepreneurs pitch at Andreessen Horowitz (a16z), a major Silicon Valley venture-capital firm, there’s a high chance their startups are running on Chinese models. “I’d say there’s an 80% chance they’re using a Chinese open-source model,” notes Martin Casado, a partner at a16z.“ —- https://ixbroker.com/blog/china-is-quietly-overtaking-americ...
crazygringo
Just searched for their actual policies to corroborate and found the policies on ID verification:
https://help.openai.com/en/articles/10910291-api-organizatio...
And that credits are nonrefundable:
https://openai.com/policies/service-credit-terms/
It absolutely seems like terrible horrible customer service not to issue refunds in this case. Obviously the credits can still be used for most of the models, so it's not like you can't do anything with them. But if someone explains they bought the credits specifically to use with the verification-gated models and then discovered they couldn't (since apparently verification fails for some people), there's no question that refunds are the right thing to do. What is OpenAI thinking?
(BTW, speculation seems to be that the verification process doesn't have anything to do with know-your-customer laws or anti-fraud, but is intended to prevent competitors like Chinese DeepSeek from having large-scale access to OpenAI's best models.)
pjmlp
Depending on where the post author is located, whatever says on those links is worth garbage if they are located in Europe.
Most European countries have consumer protection agencies with teeth, and a company cannot decide on their own what they refund or not.
retube
This may be true, but potentially involves much time, energy, and money by the author to challenge.... so for 99% of people, OpenAI will get away with it
potamic
That's kinda scammy. It's not like they have to manage shipments and handle goods or anything. I wonder if they're banking on a percentage of users leaving credits unused like credit card companies do with loyalty points.
helicone
I don't think they care one way or the other. They haven't ever been profitable, and so they're likely going to build up data and pull the rug on all of their users by suddenly declaring themselves a data broker. They won't try this against companies that can afford to sue, but most of their users will probably start to get even more creepily targeted ads directed at them.
logicchains
>(BTW, speculation seems to be that the verification process doesn't have anything to do with know-your-customer laws or anti-fraud, but is intended to prevent competitors like Chinese DeepSeek from having large-scale access to OpenAI's best models.)
It's not because OpenAI's CEO is also the founder of WorldCoin, a project to ID everyone?
egorfine
Funny though their KYC process is not done via WorldCoin. Obviously because WorldCoin KYC is useless for authorities.
irvingprime
Customer service? In the age of AI? What have you been smoking?
gdulli
It's hard to know exactly which forms it will end up taking, but dependence on these companies is going nowhere good. And more quickly than it took streaming (for example) to go from offering a better experience (to win market share) to the current (and inevitable) norm of raising prices constantly and introducing unskippable ads.
dawnerd
I could totally see them having responses sprinkled with subtle marketing. Ask it for the best travel backpack and ooops all sponsored.
conception
This is already the case for OpenAI. Go ask for some backpack recommendations.
CaptainOfCoit
> and introducing unskippable ads.
Maybe I'm missing something obvious, but where on either ChatGPT or the API platform OpenAI hosts are you seeing ads?
mapontosevenths
> it took streaming...
Parent is comparing OpenAI to other companies that followed a similar trajectory of enshittification.
gdulli
Yes, and just as ads eventually came to streaming in a worse form (unskippable, hypertargeted) that cable companies didn't have the ability to innovate, the new frontier of ads will again come with qualitatively worse innovations. Seamless and undisclosed in conversational LLM output.
jarym
Wonder how long before they'll have to start reporting 'suspicious activity' to the government same as financial institutions have to do for money transfers.
A4ET8a8uTh0_v2
You can reasonably assume it is already happening. The only difference is that for FIs it is required by law, that it is relatively similar across the board in terms of implementation and openai is a one giant source of info you wouldn't get anywhere else.
It fairly accurately measured my age, location, place of birth and political inclinations based on our conversations alone. I am certain it can infer a lot more.
egorfine
This.
There is no other reason to require KYC for a server-side text transformation tool, no matter how impressive it is.
Mars008
The other reason could be the copyright cases they are fighting in court. OAI was ordered to keep all records, including private. Not sure if it was lifted already.
And another could be EU requirements for age verification. AI can produce adult content.
There are may be other reasons, like to prevent using OAI models' output to train competing models.
weird-eye-issue
Absolutely not. It would require product, engineering, admin, etc. effort to do that and unless it isn't required by law why would they waste the time when they have a lot else to do?
bgwalter
They have an ex-NSA chief on the board, and doing surveillance voluntarily may result in government help like getting contracts in South-Korea and Argentine that may bring in far more money than the implementation costs. Perhaps they outsource the implementation to Palantir or the NSA. It is basically a simple middleware that is inserted somewhere once the traffic is decrypted.
So I don't think implementation costs are an obstacle.
orthecreedence
> why would they waste the time
Because then the NSA shows up with an NSL, you integrate with the fascist surveillance state or you lose your business. How have people forgotten this so fucking quickly?
queenkjuul
I'd have sworn they've already admitted to this
yalogin
If deepseek and qwen are capable why does one need to use OpenAI models? Are their models really that much better? If not the only they bring is the hosting service. N that scenario how long do they have this advantage before aws or Microsoft takes over?
Mars008
> If deepseek and qwen are capable why ...
Of course, you can go further and run qwen locally. Or even train your own nanogpt. Why not if it's capable, right? And this 'if' is a big question.
puppycodes
I wouldn't tie my email to a chatbot let alone my literal goverment ID.
seneca
Yeah, I'm not at all willing to do these sorts of verifications. Any company doing them essentially doesn't exist to me. I don't even use Anthropic because they require a phone number to register.
quantummagic
Same. I don't understand why so many people are happy to give their phone number to some random service provider. It's a shame it has become normalized.
CaptainOfCoit
My phone number is basically public and has been for 20 years, every email I send has my phone number and it's findable via the public internet too.
Not sure why people see their phone number as something private?
FWIW, I've heard some people saying they avoid it because of spam, I've been on my local anti-spam list since I got my current phone number, and receive about 1 spam call every week or something. Maybe there is one for where you live too.
seneca
You're lucky. In my experience no-call lists don't work.
I command a significant budget and even with a lot of effort to not proliferate my phone number, I get at least half a dozen spam or sales calls a day. I can't imagine how bad it would be if I didn't attempt to protect it. Perhaps it would be the same and I should just give up, but I'm not willing to try.
The other side of the coin is that it's just none of their business. They don't need my phone number to sell me SaaS software. There is no upside for me to give it to them.
binarymax
GPT-5 works, just not GPT-5 streaming. I posted about this a little while ago with more details: https://news.ycombinator.com/item?id=44837367
thr0w
What is it about streaming specifically that necessitates this? Am I missing something obvious?
deaux
I thought this was indeed the case at release, but then they changed it to also be for non-streaming (completions). So either they reverted it back or it was a temporary bug during the early days of the model.
When did you last check?
comrade1234
I put $2 on my deepseek account and have barely used it, it's so cheap.
egorfine
Indeed. I have opened the playground and it doesn't let me choose GPT-5. Obviously I will not be KYCing myself.
But that's okay. There are plenty of other models. Perhaps not bleeding edge great, but great nevertheless.
Just frustrated here: I credited my OpenAI API account with credits, and then it turns out I have to go through some verification process to actually use the API, which involves disclosing personal data to some third-party vendor, which I am not prepared to do. So I asked for a refund and am told that that refunds are against their policy.
So I'll be cancelling my chatgpt plus sub, disputing the card payment, and moving to deepseek.
Edit: Deepseek seems to be a lot cheaper than OpenAI
Edit 2: seems verification is only needed for gpt-5, gpt-4o seems to work without it