Recently a few days back, I had to verify my Linkedin identity on a new account (I am 17 for context) and I used proton mail and Linkedin immediately blocked it and asked for verification
I legally couldn't verify because persona doesn't detect aadhaar card and their support system on twitter/mail whatever was incredibly bad so much so that it felt like copy-paste and I still haven't gotten the card. I have written about my experience too.
https://smileplease.mataroa.blog/blog/linkedin/ : (Title of this is) Linkedin's "final decision", restricting my account, making me feel unheard, Persona being Persona & the time I asked Linkedin support what 351/13 is to prove if they are human or not.
An equally valid question is "does the company you use for identify verification follow the same commitments with regards user privacy and selling/processing of user data as Anthropic itself?".
And the answer to that question is:
"Hell no! We used the cheapest, shadiest company we could find for that. They'll process and sell all your data. Thank you for continuing to be a valued Anthropic customer!".
* preventing North Korea, China, Russian, Iran and etc. actors from accessing service. They absolutely use workarounds to access AI, e.g. I bet there are companies who are proxy between Anthropic and those countries.
I imagine there will be quite some false positives while identifying those.
This will do absolutely nothing to prevent those actors from accessing Claude... they already recruit young unemployed Americans to do proxy job interviews[0][1], etc. They'll just pay young unemployed Americans to do verification for them.
That sounds likely to increase their costs and create new opportunities to get caught. Not a silver bullet but not "absolutely nothing". Like how anti-money laundering laws don't wipe out all crime, but are still worthwhile.
Just a few days ago, on Friday, my 15 year old son had his Claude account suspended with a demand for ID to prove he is 18 or older. He had his own Claude Max subscription (he out-earns me fairly frequently in his circle of gaming programmers), and was unaware Anthropic had a must-be-18 rule, as was I. Their email said "Our team found signals that your account was used by a child. This breaks our rules, so we paused your access to Claude." So I guess if you ever ask a question that seems to originate from a teen or less, expect to hit an ID gate.
So now he's a Codex user. OpenAI and Google both have a minimum age of 13.
EDIT: I should note that Anthropic gave him a refund for the whole month that was underway, despite him being nearing the end of it. So good on them.
People have tried to run Qwen3-235B-A22B-Thinking-2507 on 4x $600 used, Nvidia 3090s with 24 GB of VRAM each (96 GB total), and while it runs, it is too slow for production grade (<8 tokens/second). So we're already at $2400 before you've purchased system memory and CPU; and it is too slow for a "Sonnet equivalent" setup yet...
You can quantize it of course, but if the idea is "as close to Sonnet as possible," then while quantized models are objectively more efficient they are sacrificing precision for it.
So next step would be in order to up the speed, we're at 4x $1300, Nvidia 5090s with 32 GB of VRAM each (128 GB), or $5,200 before RAM/CPU/etc. All of this additional cost to increase your tokens/second without lobotomizing the model. This still may not be enough.
I guess my point is: You see this conversation a LOT online. "Qwen3 can be near Sonnet!" but then when asked how, instead of giving you an answer for the true "near Sonnet" model per benchmarks, they suddenly start talking about a substantially inferior Qwen3 model that is cheap to run at home (e.g. 27B/30B quantized down to Q4/Q5).
The local models absolutely DO exist that are "near Sonnet." The hardware to actually run them is the bottleneck, and it is a HUGE financial/practical bottleneck. If you had a $10K all-in budget, it isn't actually insane for this class of model, and the sky really is the limit (again to reduce quantization and or increase tokens/second).
PS - And electricity costs are non-trivial for 4x 3090s or 4x 5090s.
Qwen3.5-35B-A3B is reported to perform slightly better than the model you mentioned.
It runs fine on a single 3090 with 131072 tokens of context (or even twice that, but I wanted some VRAM left over), and due to it being MoE, the context size scales rather less drastically than ctx^2. I've had friends with smaller cards still getting work out of it. Generation is at around 20 tokens/sec on that 3090. You'll need enough DRAM to hold the bits of the model that don't fit. Nothing to write home about, but genuinely usable in a pinch or for tasks that don't need immediate interactivity.
It's the first local model that passes my personal kimbench usability benchmark at least. Just be aware that it is extremely verbose in thinking mode. Seems to be a qwen thing.
A machine with 128GB of unified system RAM will run reasonable-fidelity quantizations (4-bit or more).
If you ever want to answer this type of question yourself, you can look at the size of the model files. Loading a model usually uses an amount of RAM around the size it occupies on disk, plus a few gigabytes for the context window.
Qwen3.5-122B-A10B is 120GB. Quantized to 4 bits it is ~70GB. You can run a 70GB model in 80GB of VRAM or 128GB of unified normal RAM.
Systems with that capability cost a small number of thousand USD to purchase new.
If you are willing to sacrifice some performance, you can take advantage of the model being a mixture-of-experts and use disk space to get by with less RAM/VRAM, but inference speed will suffer.
It seems out of step and foolish, and the cynic in me says that Anthropic has a side hustle of identity harvesting and is looking for justifications, but on the flip side, there is a real risk of pearl clutching if a child ever uses AI, and maybe Anthropic just wants to steer clear of all of that. Though simply putting it in the ToS should be sufficient legal shielding, and the idea that they're chat harvesting to age fingerprint conversations seems dubious.
Wouldn't the reasons for requesting identification be the same those for banning people - the system has flagged that you might be from the wrong location/under 18/creating multiple free acounts etc - so is validating.
They request ID for bans so that they can ban you personally. ID checks may as well be a sign that you've already been banned and they're fishing for ways to make the ban harder to evade. Venmo does the same thing.
Maybe Anthropic just likes creating a market for dark identities. Because that's the most likely effect of such stupidity; generating more ID theft victims with no change to services to criminals.
Is a "dark identity" one that's never been shared with an identity-theft-as-a-service? Or is it just of one that's (supposed to be) privacy-conscious (and wouldn't otherwise have been an easy victim)?
Persona is easy to implement, has all the compliance requirements, and is in line with market prices. ID verification will always be an afterthought, unfortunately.
Identity verification to use an API?? And via Persona? I can't say if it's real. But if they really try to enforce that, I guess goodbye Anthropic forever.
They were all the same from the beginning. Every tech company of a certain size and significance eventually begins collecting data and sharing it with state actors, as far as I can see.
In the old USSR one had to register a typewriter. Sweet memories. And at that time western people (deservedly) laughed at it or used facts like this to show how backwards the country was
"Being responsible with powerful technology starts with knowing who is using it."
In other words: they want to create a private web and sniff-after-people system. Today the EU also introduced an app for age verification. They also constantly say how this is ... voluntary.
Well, I guess we all know the direction. Let's have a look at this in a few years, because there may be a few ... suspicions.
With regards to Claude the question is: WHY do they want to sniff off user data exactly?
Ugh what a disaster. This is so Anthropic can enforce bans.
The future has arrived, in which you are only allowed to program a computer in any meaningful way requires total identification and permission.
What a tragedy that the amazing capabilities of LLM assisted programming come with such disgusting and reprehensible requirements and impositions.
So they can ban you from some minor infringement of their usage policies and you'll never be allowed to program again.
"Mr Anderson, it has come to our attention that you have been programming computers under an assumed identity. As you are aware this is a felony under the computer fraud and hacking act and you will be sentenced to four years in jail and may never use a computer again.". Yes laugh it up.
I may consider showing my ID to a company I already have a business relationship with; given demonstrable legal obligations, contractual necessities, legitimate interests etc . Eg the standard GDPR list.
I do have an existing business relationship with Anthropic, so I might under some circumstances decide to show them my id. I don't have a business relationship with Persona though.
I understand the instinct: they want to insulate themselves from holding PII. Not the worst idea. I'm not happy with it being a third party though. Especially the third party in question.
This is deranged. Say you wanted to use AI to prepare whistleblowing submission to use regulatory language and test for any weak points. Then Claude flags it and requires you to identify yourself. It's not a stretch of imagination that before you manage to send the bundle, you find yourself in the suitcase somewhere in the woods. People explore all kinds of sensitive stuff and I see it is tempting for AI companies to see exact person behind it and then it takes one disgruntled employee to put lives in danger.
WTF
>Say you wanted to use AI to prepare whistleblowing submission to use regulatory language and test for any weak points.
Why would you do this? If you can't write it yourself, you're just sabotaging your effort once the hallucinations are revealed. Secondly, a whistleblower is going to use a corporate LLM provider? Even without ID checks, that's an extremely uncompensated risk.
Persona is bad news. They should not be using Persona. This is bad.
> Your ID and selfie are collected and held by Persona, not on Anthropic's systems. Anthropic can access verification records through Persona's platform when needed—for example, to review an appeal—but we don't copy or store those images ourselves.
It's unacceptable that this data is persisted at all, let alone that it's persisted by Persona.
> Persona is contractually limited in how they can use your data: only to provide and support verification and to improve their ability to prevent fraud. They're bound to protect it with industry-standard security controls and delete it in line with the retention limits we've set and applicable law.
It's good to hear that they're criminals. That means nothing for me though. Nothing.
> Why did my account get banned after verification?
This is bad. Why do they wait to ban until after they have your personal info? Venmo did the same thing to me: They didn't tell me I was banned until they had my ID. Absolutely despicable practice.
---
Anthropic is one of my favorite AI companies because they get LLMs more right than anyone else I've seen. But unfortunately this also means they can be swindled by social manipulation in lieu of technical excellence; the same type of brain results in both, I've seen it.
Persona is a bout of sociopaths, and it shows: they're worming their way into everything despite the well-documented conspiracy. They're doing it out in the open with zero consequences.
Why is this necessary if I'm paying Anthropic with a credit card? A credit card requires a) credit worthiness, b) a line of credit from a bank based on the individual's identity, and c) regular payments. Why isn't a credit card enough? Why can't certain features be paywalled?
If someone is doing something deeply unethical with Claude, let's say they're using a clade of Claudes to launch cyberattacks, then doesn't Anthropic have fine grained telemetry, payment history, API usage / prompting / requests, and other details necessary to investigate? What does a government photo ID provide Anthropic that these data points don't?
At this point, people usually ask "what if they use stolen credit cards?" or are "state backed?" then well... if they're state backed / using stolen credit cards, then they're also capable of using stolen IDs or state-sponsored "legitimate" IDs.
It doesn't make much of a difference to organized crime / state backed assets. Or, Anthropic. But it makes A HUGE difference for entrepreneurs, founders, and just plain old consumers who use the service.
It's an asymmetric risk.
It's one thing for your credit card to leak, you can get a new one. It's harder for lower-tier / dumber criminals to socially engineer into your personal information for impersonation / ID theft with just a credit card number. But it becomes a lot easier with your scans of your ID.
Unless you're connected with an org of interest, have b/millions in crypto, most better organized groups / state actors won't usually (no guarantees) steal your identity. Identity theft is very much a SME operation in cybercrime.
So when Persona inevitably gets compromised and everyone's personal IDs inevitably gets leaked, the threat posed to entrepreneurs, founders and consumers is higher than the inverse.
I don't understand why Anthropic would expose themselves to the liability; when arguably they have all the tools baked right in.
I don't use their tool for writing. Perhaps it's ego, but I think I'm a better writer. But I shared the above text and asked Claude Opus 4.6 on Max thinking, "What would you say about the argument that the Anthropic has the best tool for threat prevention baked right in?"
Claude is the threat prevention. It's sitting between every user and every potential misuse, in real-time, at every interaction. It refuses harmful requests. It detects prompt injection. It flags dangerous patterns. Anthropic has built the most sophisticated content-aware security layer in history — and it operates at the interaction level, where misuse actually happens.
A JPEG of someone's driver's license sitting in a Persona database does exactly nothing at the moment someone tries to use Claude for harm. Claude's own refusal system does everything.
So the full argument stacks:
1. Credit cards already verify identity (bank KYC)
2. Anthropic's telemetry already detects misuse patterns better than any static document
3. The AI itself is the security layer — real-time, context-aware, at the exact point of interaction
4. Photo ID adds zero marginal security — while concentrating breach risk on users
Three layers of existing protection, all superior to a photo ID. The ID is the weakest link in the security model and the highest-risk data asset in the system. It's the only component that, when breached, harms the user more than the company.
You should write this up.
Persona collects far more info on you than a name and credit card number. There are also some obfuscation services for cards (will it reject a prepaid debit card?), which would require them to go through extra steps to get your actual identity.
>I don't understand why Anthropic would expose themselves to the liability; when arguably they have all the tools baked right in.
What liability? When has a company ever faced any significant penalty for irresponsibly handling people's private data?
> Why is this necessary if I'm paying Anthropic with a credit card?
You can have a CC / Visa / MasterCard when you are under 18 years old, but you need to be 18 or older for Claude. That would be one reasons why CC does not work.
Or maybe they suspect you opened a second account after your first got banned for whatever reason. Like you said it's easy to get a new card.
Persona also might send your data to 17 different subprocessors (16 if you exclude Anthropic itself).
You reminded me of this submission from two months ago: I verified my LinkedIn identity. Here's what I handed over (https://news.ycombinator.com/item?id=47098245)
I legally couldn't verify because persona doesn't detect aadhaar card and their support system on twitter/mail whatever was incredibly bad so much so that it felt like copy-paste and I still haven't gotten the card. I have written about my experience too.
https://smileplease.mataroa.blog/blog/linkedin/ : (Title of this is) Linkedin's "final decision", restricting my account, making me feel unheard, Persona being Persona & the time I asked Linkedin support what 351/13 is to prove if they are human or not.
And the answer to that question is:
"Hell no! We used the cheapest, shadiest company we could find for that. They'll process and sell all your data. Thank you for continuing to be a valued Anthropic customer!".
* preventing North Korea, China, Russian, Iran and etc. actors from accessing service. They absolutely use workarounds to access AI, e.g. I bet there are companies who are proxy between Anthropic and those countries.
I imagine there will be quite some false positives while identifying those.
[0] https://www.tradingview.com/news/cointelegraph:6192f38e3094b...
[1] https://youtube.com/watch?v=QebpXFM1ha0
So now he's a Codex user. OpenAI and Google both have a minimum age of 13.
EDIT: I should note that Anthropic gave him a refund for the whole month that was underway, despite him being nearing the end of it. So good on them.
It is frequently said that programming directly is obsolete, and the skill you must have now is knowing how to operate agentic AIs.
Yet you aren't allowed to do this until you're 18.
So, developing software is now 18+ only?
Local models are chasing the online frontier models pretty hard.
So worst case, that's the fallback (FWIW, YMMV)
Gemma 4 31B Q6: 9tok/s, I'd say it is smarter than GPT-4o, but yeah it's slow. Good for coding.
Gemma 4 26B A4B Q4: 50tok/s. Feels faster than ChatGPT 5.4, but not as smart (as it reasons less). Good for general chatting and research.
You can quantize it of course, but if the idea is "as close to Sonnet as possible," then while quantized models are objectively more efficient they are sacrificing precision for it.
So next step would be in order to up the speed, we're at 4x $1300, Nvidia 5090s with 32 GB of VRAM each (128 GB), or $5,200 before RAM/CPU/etc. All of this additional cost to increase your tokens/second without lobotomizing the model. This still may not be enough.
I guess my point is: You see this conversation a LOT online. "Qwen3 can be near Sonnet!" but then when asked how, instead of giving you an answer for the true "near Sonnet" model per benchmarks, they suddenly start talking about a substantially inferior Qwen3 model that is cheap to run at home (e.g. 27B/30B quantized down to Q4/Q5).
The local models absolutely DO exist that are "near Sonnet." The hardware to actually run them is the bottleneck, and it is a HUGE financial/practical bottleneck. If you had a $10K all-in budget, it isn't actually insane for this class of model, and the sky really is the limit (again to reduce quantization and or increase tokens/second).
PS - And electricity costs are non-trivial for 4x 3090s or 4x 5090s.
Qwen3.5-35B-A3B is reported to perform slightly better than the model you mentioned.
It runs fine on a single 3090 with 131072 tokens of context (or even twice that, but I wanted some VRAM left over), and due to it being MoE, the context size scales rather less drastically than ctx^2. I've had friends with smaller cards still getting work out of it. Generation is at around 20 tokens/sec on that 3090. You'll need enough DRAM to hold the bits of the model that don't fit. Nothing to write home about, but genuinely usable in a pinch or for tasks that don't need immediate interactivity.
It's the first local model that passes my personal kimbench usability benchmark at least. Just be aware that it is extremely verbose in thinking mode. Seems to be a qwen thing.
If you ever want to answer this type of question yourself, you can look at the size of the model files. Loading a model usually uses an amount of RAM around the size it occupies on disk, plus a few gigabytes for the context window.
Qwen3.5-122B-A10B is 120GB. Quantized to 4 bits it is ~70GB. You can run a 70GB model in 80GB of VRAM or 128GB of unified normal RAM.
Systems with that capability cost a small number of thousand USD to purchase new.
If you are willing to sacrifice some performance, you can take advantage of the model being a mixture-of-experts and use disk space to get by with less RAM/VRAM, but inference speed will suffer.
Who says this?
- Repeated violations of our Usage Policy
- Account creation from an unsupported location
- Terms of Service violations
- Under-18 usage
For the user, sure. But for companies and governments? I'm pretty sure Person is quite trustworthy.
In other words: they want to create a private web and sniff-after-people system. Today the EU also introduced an app for age verification. They also constantly say how this is ... voluntary.
Well, I guess we all know the direction. Let's have a look at this in a few years, because there may be a few ... suspicions.
With regards to Claude the question is: WHY do they want to sniff off user data exactly?
The future has arrived, in which you are only allowed to program a computer in any meaningful way requires total identification and permission.
What a tragedy that the amazing capabilities of LLM assisted programming come with such disgusting and reprehensible requirements and impositions.
So they can ban you from some minor infringement of their usage policies and you'll never be allowed to program again.
"Mr Anderson, it has come to our attention that you have been programming computers under an assumed identity. As you are aware this is a felony under the computer fraud and hacking act and you will be sentenced to four years in jail and may never use a computer again.". Yes laugh it up.
I may consider showing my ID to a company I already have a business relationship with; given demonstrable legal obligations, contractual necessities, legitimate interests etc . Eg the standard GDPR list.
I do have an existing business relationship with Anthropic, so I might under some circumstances decide to show them my id. I don't have a business relationship with Persona though.
I understand the instinct: they want to insulate themselves from holding PII. Not the worst idea. I'm not happy with it being a third party though. Especially the third party in question.
Why would you do this? If you can't write it yourself, you're just sabotaging your effort once the hallucinations are revealed. Secondly, a whistleblower is going to use a corporate LLM provider? Even without ID checks, that's an extremely uncompensated risk.
> Your ID and selfie are collected and held by Persona, not on Anthropic's systems. Anthropic can access verification records through Persona's platform when needed—for example, to review an appeal—but we don't copy or store those images ourselves.
It's unacceptable that this data is persisted at all, let alone that it's persisted by Persona.
> Persona is contractually limited in how they can use your data: only to provide and support verification and to improve their ability to prevent fraud. They're bound to protect it with industry-standard security controls and delete it in line with the retention limits we've set and applicable law.
It's good to hear that they're criminals. That means nothing for me though. Nothing.
> Why did my account get banned after verification?
This is bad. Why do they wait to ban until after they have your personal info? Venmo did the same thing to me: They didn't tell me I was banned until they had my ID. Absolutely despicable practice.
---
Anthropic is one of my favorite AI companies because they get LLMs more right than anyone else I've seen. But unfortunately this also means they can be swindled by social manipulation in lieu of technical excellence; the same type of brain results in both, I've seen it.
Persona is a bout of sociopaths, and it shows: they're worming their way into everything despite the well-documented conspiracy. They're doing it out in the open with zero consequences.
If someone is doing something deeply unethical with Claude, let's say they're using a clade of Claudes to launch cyberattacks, then doesn't Anthropic have fine grained telemetry, payment history, API usage / prompting / requests, and other details necessary to investigate? What does a government photo ID provide Anthropic that these data points don't?
At this point, people usually ask "what if they use stolen credit cards?" or are "state backed?" then well... if they're state backed / using stolen credit cards, then they're also capable of using stolen IDs or state-sponsored "legitimate" IDs.
It doesn't make much of a difference to organized crime / state backed assets. Or, Anthropic. But it makes A HUGE difference for entrepreneurs, founders, and just plain old consumers who use the service.
It's an asymmetric risk.
It's one thing for your credit card to leak, you can get a new one. It's harder for lower-tier / dumber criminals to socially engineer into your personal information for impersonation / ID theft with just a credit card number. But it becomes a lot easier with your scans of your ID.
Unless you're connected with an org of interest, have b/millions in crypto, most better organized groups / state actors won't usually (no guarantees) steal your identity. Identity theft is very much a SME operation in cybercrime.
So when Persona inevitably gets compromised and everyone's personal IDs inevitably gets leaked, the threat posed to entrepreneurs, founders and consumers is higher than the inverse.
I don't understand why Anthropic would expose themselves to the liability; when arguably they have all the tools baked right in.
I don't use their tool for writing. Perhaps it's ego, but I think I'm a better writer. But I shared the above text and asked Claude Opus 4.6 on Max thinking, "What would you say about the argument that the Anthropic has the best tool for threat prevention baked right in?"
(I did.)>I don't understand why Anthropic would expose themselves to the liability; when arguably they have all the tools baked right in.
What liability? When has a company ever faced any significant penalty for irresponsibly handling people's private data?
You can have a CC / Visa / MasterCard when you are under 18 years old, but you need to be 18 or older for Claude. That would be one reasons why CC does not work.
Or maybe they suspect you opened a second account after your first got banned for whatever reason. Like you said it's easy to get a new card.