Anthropic adds ID verification to Claude via Persona

My notes

Summary

Anthropic is rolling out identity verification for select Claude use cases, using Persona Identities as the third-party verifier. Users may be asked to upload a government photo ID and take a live selfie; the data is held by Persona (not Anthropic) and is contractually walled off from model training, marketing, or third-party sharing.

Key Insight

• Verification is conditional, not blanket, it’s triggered for certain capabilities, “routine platform integrity checks,” and safety/compliance reasons. Expect it on higher-risk features (Computer Use, agents, etc.).
• Vendor: Persona Identities. Anthropic is the data controller; Persona is the processor. ID + selfie sit on Persona’s infrastructure; Anthropic can pull verification records via Persona’s platform (e.g. for appeals) but doesn’t copy/store them.
• Hard “no” list for ID type: photocopies, screenshots, scans, mobile/digital IDs (incl. mDLs), student/employee/library/bank cards, temporary paper IDs. Only physical, government-issued, undamaged, legible photo IDs pass. This rules out the increasingly common mobile driver’s license in US states.
• Explicit promises worth noting in privacy/AI policies that build on Claude:
  • Verification data not used to train models.
  • Data minimisation: only what’s needed for the check.
  • No third-party sharing for marketing/ads, only legal process.
  • Encryption in transit and at rest.
• Ban path is separate from verification failure: even after verifying, accounts can be terminated for Usage Policy violations, unsupported location, ToS breaches, or under-18 usage. Appeal lives at claude.ai (must be logged in to the banned account).
• Failure modes are mundane: blurry photo, unreadable doc, expired ID, technical glitch. Multiple retries allowed inside the flow before the appeal form (claude.com/form/identity-verification-help) is the only path.