Identity Verification on Claude: Hands-On Guide (2026)

The end state you’re aiming for: a green checkmark from Persona, no pop-up the next time you open Claude, and your workflow uninterrupted. That’s it. Five minutes of camera work and you’re back to whatever you were doing.

The catch is that identity verification on Claude just dropped as a formal policy update – emails went out around June 10, 2026, with the rule kicking in July 8 – and a lot of people are getting tripped up by failure modes nobody bothered to document. So we’re going to walk backwards from the finish line: what the verified state looks like, then the exact click path, then the three ways it silently fails.

What this verification actually is (briefly)

A live ID check. You point your phone or webcam at a government-issued photo ID, then at your own face, and Persona Identities – Anthropic’s KYC vendor – confirms the two match. Anthropic’s help center notes Persona was chosen based on the strength of their technology, privacy controls, and security safeguards. Your ID images and selfie sit on Persona’s servers – Anthropic accesses verification records through Persona’s platform when needed, but doesn’t copy or store the images itself.

Who sees the prompt: Free, Pro, and Max consumer subscribers. Team, Enterprise, and API plans are exempt. That exemption matters more than people realize, and we’ll come back to it.

The walkthrough, working backwards

You’ll know you need to do this when a Persona-branded modal appears mid-session, usually after a login or when you try to use a higher-trust feature. Here’s the flow in reverse – the last step first, so you know what you’re aiming at.

1. Final screen: “Verification complete.” Close the modal. Claude resumes where you left off. No email confirmation usually – the session itself unlocks.
2. Selfie capture: Persona’s webcam or phone camera prompt asks for a live selfie. Not an upload – a live capture. Hold still, neutral expression, decent lighting.
3. Document capture: Front of the ID, then back if it has one (driver’s licenses do, passports don’t). Persona reads the MRZ or barcode, so glare on the lamination is the #1 reason this stalls.
4. Document type selection: Choose passport, driver’s license, national ID, or state/provincial ID. Student IDs, employee badges, library cards, and bank cards are not accepted.
5. Country selection: Pick the country that issued the document, not the country you’re sitting in. Mismatches don’t block you, but they slow the auto-detection.
6. Trigger: A modal appears inside Claude. You didn’t click anything to start this – it just showed up.

Per the Claude Help Center, total time is typically under five minutes if nothing goes wrong. We’re about to talk about what goes wrong.

The three failure modes nobody documents

This is where most tutorials stop and most readers get stuck. Verification can fail because of poor lighting, an unclear image, or an expired document – and the modal usually just says “try again” without telling you which one.

Lighting. Persona’s liveness check needs even light on your face. Overhead office light works. A window behind you doesn’t – you become a silhouette. Phone flashlight in a dark room is worse, not better; it creates hot spots that the algorithm reads as a printed photo held up to the camera.

Expired ID. This one catches international users hard. A passport that expired three weeks ago will scan fine – the OCR reads the document – but the expiry date kicks it out at the validation step. The error wording doesn’t always say “expired”; sometimes it reads “document could not be verified.” Check the date before you start.

Digital or photocopied ID. Photocopies, screenshots, digital IDs (including Apple Wallet driver’s licenses in supported US states), and temporary paper IDs are all rejected. You need the physical card in your hand. The system is good at detecting screen reflections, so holding your phone up to your laptop camera won’t work.

Before you start: do a 10-second dry run. Open your phone camera, point it at your ID under your actual lighting, and check if you can read every character including the expiry date. If you can’t, Persona can’t either. Fix the lighting before you open the verification modal – attempts are capped, and exhausting them means a support-form queue, not a retry button.

What if you’d rather not hand over biometric data?

There’s an escape hatch most coverage misses. Switching to Claude Team, Enterprise, or API tiers removes you from the consumer policy entirely. The API in particular – same models, billed per token, no consumer ID gate. If you’re a developer using Claude inside a script or a tool like Cursor, you may already be on the API and never see a prompt at all.

API access means giving up Projects, Artifacts UI, and the claude.ai chat interface unless you build your own. For a casual chat user, that’s a real downgrade. Developers running automations, though? They probably should have been on the API all along – the chat interface was never the right surface for scripted workflows anyway.

How Claude compares to ChatGPT and Gemini

Quick reality check: the framing in some headlines makes this sound like Anthropic is uniquely strict. It is – but the gap is worth measuring.

Among the three major consumer chatbot platforms, Claude is the only one with a formal biometric KYC step at the consumer tier as of this writing. The other labs may follow – they usually do – but right now this is unique to Claude.

That raises a question worth sitting with: if identity verification becomes table stakes for AI access, what does that mean for users in countries where government IDs are tied to political risk? Anthropic’s policy doesn’t address this yet, and the answer will matter for how the broader industry develops the practice.

One thing the docs don’t answer

What actually triggers a check? Nobody outside Anthropic knows. The official team has not provided specific triggering conditions; more detail is likely to surface after July 8, per KuCoin’s June 2026 reporting. Community reports point to new accounts, unusual login geography, and high-volume agent activity – but none of that is confirmed. If you’re running Claude inside a long-lived automation, plan for the modal to appear at the worst possible moment. That’s the only safe assumption.

And one thing worth holding onto: verifying doesn’t make your account permanently safe. Passing verification does not guarantee that an account will never be restricted – accounts can still be disabled for other safety-process reasons. Verification is a gate, not a shield.

FAQ

Does Anthropic train its models on my ID photo?

No. Per the Claude Help Center, identity data is not used to train models – it’s used solely to confirm who you are and meet legal and safety obligations.

I’m a developer running Claude inside an automation. Will my agent break when this lands?

Depends entirely on which API surface you’re using. Scripts hitting api.anthropic.com with a Console API key are on the developer/business path – the consumer ID check doesn’t apply. If your “agent” is actually a browser-automation thing that logs into claude.ai with email and password, then yes, you’ll hit the modal. Turns out the selfie step is intentionally human-only, with no headless path through it. Move that workload to the API before July 8 if you can.

What happens if I just refuse?

You lose access to whatever feature triggered the prompt. No hard account-deletion timeline for refusal has been published, and trigger conditions are still opaque – so practically speaking, you’re locked out of the gated feature until you verify or switch to a tier that doesn’t require it.

Next step: before July 8, check your physical wallet. If your government ID is within six months of expiring, renew it now – a fresh document is the single biggest factor in a clean first-attempt verification.