More reasons to go local: Claude is beginning to require identity verification, including an valid ID like passport or drivers license and a facial recognition scan.
Posted by fulgencio_batista@reddit | LocalLLaMA | View on Reddit | 72 comments
Makers7886@reddit
Wonder how much of this is part of the "US Labs binding together to stop Chinese labs from using them" or how much is just an excuse to extract personal data.
More-Curious816@reddit
It's not about stopping the Chinese labs. You can't stop it, with money you can start proxy companies, teams and get that sweet data via the apis for corporations use. Literally trying to hold water in a sieve. Also they can't stop the Chinese labs because their government actually think AI is important for their sovereignty. Plus, china currently has the most AI papers outputs, citations, AI related patents, and robotics labs. American labs are already behind according to latest Stanford University report.
It's just personal informations for ads, always ads.
doodlinghearsay@reddit
They don't want to stop them overall, they just want them to stop doing business in the US. Same reason why you can't buy Chinese EVs in the US: American companies generally can't compete on a level playing field.
SkyFeistyLlama8@reddit
To be fair, Chinese EVs don't compete on a level playing field either. The Chinese state has massive subsidies for the entire EV supply chain from batteries to export marketing assistance. Europe slapped huge tariffs on Chinese EVs for the same reason.
Ironically, on AI, both the US and China heavily subsidize model makers and most companies in the same vertical.
xienze@reddit
Of course, but I think there’s a healthy dose of being able to tie chat history to an actual person during criminal proceedings. See: how Google search history is a standard part of investigations.
-p-e-w-@reddit
It’s neither. Anthropic leadership just has a God complex where they are honestly convinced that they have created an entirely separate class of technology that is too dangerous for regular people to access, rather than a coding bot that’s about 3 months ahead of its Chinese open competition.
Such delusions are very common in Silicon Valley and can be self-reinforcing when people are only surrounded by like minds.
drallcom3@reddit
They're just trying to profit from being the top dog. Nothing wrong with that. It's good for everyone. The general audience gets a reality check just how expensive AI is. Anyone trying to build a business around online AI providers won't ruin themselves trying without a good idea. And finally local AI models look much better in comparison.
a_mimsy_borogove@reddit
This is a great idea if their aim is to make their customers switch to Chinese models instead.
TheIncarnated@reddit
Especially with GLM competing with Claude
sammybeta@reddit
I think it's neither. They just want most real payments.
JeffDunham911@reddit
'Protecting kids', but with a fresh twist
Pretty_Challenge_634@reddit
Both can be true
En-tro-py@reddit
Porque no los dos?
h4ck3r_n4m3@reddit
This is likely going to be related to getting access to mythos or other security related models. OpenAI already started doing this
0rbit0n@reddit
did they? I use Chat GPT Pro every day and never seen it asking me for an ID... It knows who I am anyway
brainmydamage@reddit
"Or when (notice we didn't say 'if') someone in the chain inevitably gets compromised by someone with basic email and social skills."
Lissanro@reddit
A possibility thaclosed model access can be denied or that they can shutdown or change models at any time, were some of the reasons besides privacy why I went fully local, so I could run any model I need like Kimi K2.5, GLM 5.1 or Qwen3.5 397B, each has its own strengths (Qwen3.5 397B is the fastest, GLM 5.1 is the smartest but slowest because it tends to think much more has more active parameters, and K2.5 is somewhere in between).
Mochila-Mochila@reddit
Waiting to do just that once hardware has both improved and decreased in price. 1TB/1TB would be a dream, but that's for the next decade. I'd settle for 800GB/s and 256GB capacity @ 3k€, hopefully in 5 years or so.
hyperdynesystems@reddit
"Effective altruist" cult crazies strike again.
solestri@reddit
Once I learned that Anthropic was involved with that ideology, so much of their weirdness made sense.
hideo_kuze_@reddit
You now need to submit your passport and a dna sample for every fucking website or app.
How the fuck did we reach this point?
The world is becoming increasingly dystopian.
And the "freedom loving" western nations are becoming worse than supposed autocracies
FourSquash@reddit
I have not had to do this for any app whatsoever
toothpastespiders@reddit
The situation with cloud models might have improved since this happened. But I'm always going to be salty about the time I had to switch from American models to Chinese ones to work on a project related to American history.
More-Curious816@reddit
When haramby get killed. That the only possible explanation. It's the darkest timeline.
Sydorovich@reddit
Digital gulag. Welcome NWO.
vex_humanssucks@reddit
This is exactly why having capable local alternatives matters. Not anti-cloud — cloud models are great — but friction and policy changes on API access are real risks for anyone building production systems. Local inference as a fallback or primary is looking smarter every quarter.
TechnoByte_@reddit
1 day old account, clearly LLM written comments, I smell a bot
NandaVegg@reddit
OpenAI does this since o3 and it is a pure marketing gimmick so far (it is just Palantir and same entity can do verification again and again without any real human involvement). They had a broken verification for Codex they didn't bother to fix for weeks as well.
Revolutionalredstone@reddit
Agreed it's vert unsafe, I've seen lists of basically all online identity documents for very very cheap.
The chance that you're data gets leaked thru these is WELL > 90%.
Due-Function-4877@reddit
KYC for matrix math token prediction, eh? And, a unified bulletin board forum named reddit that's owned by Google to discuss things. What could possibly go wrong?
avbrodie@reddit
“Math token prediction” undersells the possible harm that can be caused by a significantly powerful LLM.
a_beautiful_rhind@reddit
What about the harm that can be caused through my brain and the local library?
avbrodie@reddit
It’s trivial to plan and execute cyber attacks with a SOA LLM; it doesn’t even need jail breaking in the traditional sense if you are sufficiently motivated.
I’m not saying that I agree with KYC for LLMs, but to say “KYC for math token prediction” is like saying “KYC for encryption keys” when discussing bitcoin. It focuses on the underlying technology without appreciating its applications.
TechnoByte_@reddit
Ah yes lemme just go to chatgpt and ask "yo find a 0-day and hack nasa"
Sure you could vibecode some malware but LLMs are useless for helping you actually get it into your target
En-tro-py@reddit
You got a license for that GPU sonny?
Sucks to be a pleb and have no data rights, if only I was born a real person registered in Delaware!
export_tank_harmful@reddit
You joke, but I wouldn't be surprised if there were a database somewhere that logged people who own a high powered GPU(s) and have accessed HuggingFace.
draconic_tongue@reddit
woah, calm down there dario
avbrodie@reddit
I chuckled lol
IrisColt@reddit
no, Google does not own Reddit, heh
Revolutionalredstone@reddit
Reminder: Everyone I know who has identity exposure online has been hacked.
My friend who uses Claude lost 100,000$ (100k) the same day he did 'verification' (sure enough he found his details were on breach forum within a few hours)
Governments and companies know this is totally unsafe and are wiling to let you get hacked.
Even important services like AI coding must be avoided if they try to force unsafe measures on it's users (like online identity exposure)
These evil draconian practices will only help push people off these 'surveillance services' and only cheaper faster Chinese models and or local full-control LLM models.
dingo_xd@reddit
So claude isn't smart enough to figure out my age by the way I speak/write?
privatetudor@reddit
I mean I wouldn't trust it to do that. Or any llm.
Having said that the day they ask for my ID is the day I cancel my plan.
TechnoByte_@reddit
Anthropic is and always has been a shitty company, there's no reason not to cancel right now
More-Curious816@reddit
If you pay money for it, can't they get the basics information and more importantly that you are an adult? Like wtf these companies after?
Zanion@reddit
It's only smart enough to hack the planet man. They aren't miracle workers.
Baphaddon@reddit
Thank god we got the python/rust releases for Claude code though
Baphaddon@reddit
PLEASE SAVE ME PRESIDENT QWEN
Ambitious_Ad4397@reddit
I think at this point we need to invest in companies that make local running LLMs more available/affordable/effective
WithoutReason1729@reddit
Your post is getting popular and we just featured it on our Discord! Come check it out!
You've also been given a special flair for your contribution. We appreciate your post!
I am a bot and this action was performed automatically.
Expensive-Aerie-2479@reddit
The closed model dependency problem hits even harder for minority languages. I've been building a Somali voice agent — Somali has ~25M speakers and there's zero support in any commercial TTS or LLM provider. ElevenLabs, Cartesia, OpenAI — nothing.
Going local with XTTS fine-tuned on ~300 hours of Somali data is the only viable path. The moment a closed API decides your language isn't worth supporting (or requires identity verification you can't provide), you're blocked entirely. Local models are genuinely the only option for a lot of the world.
Queasy_Asparagus69@reddit
Local is the future - I hope Apple figures it out, they best positioned to do something about it
cryptofriday@reddit
Local is the future !!!
a_beautiful_rhind@reddit
So I double can't have an account. First it was the VPN and now it's some face scanning humiliation ritual.
cutebluedragongirl@reddit
The beatings will continue until morale improves.
jazir55@reddit
Lmfao how much paint are they huffing? Good fucking luck getting me to do that, I'll just go with one of the 20+ other model providers. They can get gigafucked.
Ell2509@reddit
Glad I just got minimax m2.7 working on my desktop beast.
The better these 15 - 38b models get better, the faster my laptop grows in capability.
yami_no_ko@reddit
That was just a matter of time.
Luckily irrelevant, because we have a myriad of local options.
brakx@reddit
It’s so china stops stealing their model data.
Zanion@reddit
Because ofc China doesn't have the sophistication to have embedded people with valid ID.
fallingdowndizzyvr@reddit
You mean the model data they stole from everyone else?
brakx@reddit
Yes that data
fallingdowndizzyvr@reddit
LOL. So the thieves want to protect their stolen data.
substandard-tech@reddit
It’s odd that the real reason is controversial
Unable-Jelly6228@reddit
I need more 4b or 9b models lol.
One_Whole_9927@reddit
FYI It’s looking like they’re moving to lock down open source by making the US AI providers the single source of truth. If you follow the timeline of events it’s pointing to a scenario in which these fucks lock down the internet and social media.
If these companies cannot get people to subscribe to their bullshit they’re going to force it by making it impossible to use anything else. Go open source while you still can. Or look at quitgpt.org if you want more info.
a_beautiful_rhind@reddit
Oh it's literally the whole internet. Upload your ID, turn your head and cough for absolutely everything. In some cases even the device. Verify your "age" to use this ipad.
BatPlack@reddit
For the lazy:
https://quitgpt.org/
alexx_kidd@reddit
That’s ok, unless you’re an American
elongated_argonian@reddit
Nope, it applies to non-Americans too, per the link in the post.
ProdoRock@reddit
Simple solution: close your account and use another service.
xXprayerwarrior69Xx@reddit
put shoe on head
jwpbe@reddit
Please drink verification can to continue.