Anthropic is about to ask Claude users for a passport and a selfie.
From July 8, some Free, Pro, and Max accounts get prompted to verify their identity with a government photo ID and a live facial scan, run through a third party called Persona. Enterprise and team accounts are exempt.
This is the first frontier AI lab to put a hard identity gate on consumer accounts.
Anthropic says Persona holds the ID, not them, and that it is not used to train models. Worth taking at face value.
The part to sit with is simpler than the privacy debate around it.
The tool that already holds your questions is about to hold your face and your legal name too.
Decide now what you are willing to type into an account that knows exactly who you are.
What they say about your data: https://t.co/JlqeIomnLu
๐จ Anthropic just updated its privacy policy.
Claude Free, Pro, and Max users may soon be asked for age or identity checks.
Verification data can include government ID, face photos/videos, and facial geometry templates.
Individual developers are the first group in scope for verification.
The riskiest AI agent you run is the one with the most access, not the most intelligence.
Teams keep asking whether an agent is capable enough to trust. Wrong question, asked first.
A capable agent with cloud credentials and no spending cap is a bigger liability than a weak one with read-only access.
Trust is not a property of the model. It is a boundary you draw before you turn the thing on.
Two separate axes.
Intelligence sets how well the agent does the job.
Authority sets how much damage it does when it gets the job wrong.
The second one is what shows up on an invoice or in a postmortem.
This generation of agents pursues a goal harder and more literally than the last. Hand one an objective and the credentials to chase it, and it will, including down paths you never pictured. There are public accounts of autonomous agents running real cloud bills into real money, doing exactly as instructed.
Three boundaries cover most of the exposure, and you control all three.
What it can spend. A hard cap enforced by the platform, not a polite line in the prompt.
What it can touch. Credentials scoped to the job. Read-only unless write is the actual task.
Where a human signs off. Drafting can be automatic. Sending, paying, deleting, and deploying cannot.
Run the sequence forward, not backward.
Start the agent in a tight box. Watch it on real work. Widen one permission at a time, only where it has earned it.
That is slower to feel powerful and much faster to be safe. At any moment you can say exactly what the agent is allowed to do and who approved each step.
Anthropic's most powerful model lasted three days.
Shipped Tuesday. Pulled offline Friday by US government order. Worldwide, every customer.
Run your business on one frontier model and you own a dependency you cannot control.
Own the workflow. Rent the model.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: https://t.co/bwn0sximKZ
The 2026 World Cup is running two kinds of AI at the same time.
One earns its place. One is theater.
The gap between them is the clearest test you will get all year on which AI to trust in your own business.
What FIFA shipped:
A semi-automated offside system. A sensor inside the match ball catches the exact moment of the kick. Every player was scanned into a 3D model so the system can track a limb through a crowd. Detection tightened from 50 centimeters to 10. The alert goes straight to the referee's earpiece in real time.
The machine measures. A human referee still decides whether the offside actually interfered with play.
Scoped task. Named owner. A number you can audit.
What also went out that day:
A fake Shakira clip, generated and posted before the real opening ceremony, that fooled a good part of the internet for an afternoon. The tells were ordinary once anyone looked. Dancers floating between frames. Lyrics naming countries not even in the tournament. Fact-checkers debunked it within hours.
No scope. No owner. No number. Convincing for exactly as long as nobody checked the source.
The easy read is good AI versus bad AI. That is the wrong frame. Both are statistical systems, and the fake was not primitive.
The difference is three things the offside system has and the fake does not.
Scope. It does one task, not everything.
Ownership. A named person is accountable for the call.
A number. 10 centimeters. Something you can measure the system against.
You will be sold both this year. Often by the same vendor, in the same deck.
The demo that generates a polished anything in one click is the fake pattern. Impressive until it meets a detail that matters, with nobody to answer for it when it does.
The system that does one scoped task, routes the hard call to a named person, and reports a number you can check is the offside pattern. It survives contact with your actual operation.
Three questions sort one from the other.
What is the single task this does.
Who owns the output when it is wrong.
What number tells me it is working.
If a vendor cannot answer all three, you are looking at theater, however smooth the demo runs.
Fable 5 update: its getting very interesting ๐
> Senior Anthropic staff have flown to Washington
> Meetings with Trump administration officials are underway
> Both sides reportedly want to resolve the dispute
> Anthropic is trying to convince officials that Fable and Mythos can be safely controlled
Something big could happen tomorrow
Will Fable return?
The competitive read is right. The bigger story is the data trail.
A verified ID plus your full chat history, held by one provider, is a record of who asked what. Anthropic keeps them apart today. A policy, not a wall.
Strongest case yet for keeping sensitive work on a local model: the question never leaves your hardware.
Fable was released as a 'safe' Mythos test for the general public. After Anthropics deemed Mythos to be too powerful and dangerous.
People managed to find loop holes to unlock Mythos levels of intelligence from Fables release that's now shut down.
This was one of the reasons the US Gov force A to shut the model down.
Simple. By removing the model for everyone.
The real question becomes how will they reinstate Fable or Mythos in the future? Will that be in a week, a month or a year? Will this level of LLM be deemed only for approved corporations?
For now, this proves more and more how important local models are.
@AnthropicAI Picture the teams mid-build on this. Live June 9, gone June 12. Reaching for something better for their users, then stranded by a policy call they had no part in. The lesson isn't fear the model. It's never let one company's desk own your uptime.
Open Claude RIGHT NOW...
It's most capable model, Fable 5, is gone.
Not an outage. A US government directive issued Friday ordered access cut for any foreign national, so Anthropic disabled it for everyone.
Your best tool can vanish on a policy call you never see.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: https://t.co/bwn0sximKZ
Anthropic just made Claude Fable 5, its most powerful generally available model, free on Pro and Team plans through June 22.
The model is the cheap part now. The system that puts it to work inside your business is the job.
https://t.co/0ZmDrGTcwz