Anthropic Mandates Persona Biometric Verification for Claude Users

What Anthropic Changed in Its Privacy Policy

Anthropic restructured its privacy policy and notified users that it will request biometric verification to access the Claude chatbot under "certain circumstances." The company tapped identity platform Persona to handle the work: document scanning and facial-geometry calculations that confirm a person matches their credential.

The mechanics are heavy. Flagged accounts must submit high-resolution snapshots of a government credential alongside live photos or video. From those captures, the system generates biometric templates that cannot be reset. A password can be rotated after a breach; a facial-geometry template cannot, which is precisely what makes this category of data sensitive.

For most users the gate stays invisible. But once an account trips a risk signal, the path forward runs through a manual review queue and a one-way collection of irreversible identifiers.

Why the Move Drew Public Friction

The announcement drew immediate pushback from developers and privacy collectives. The objection is not that platforms should verify users, but that storing unresetable templates concentrates risk in a single vendor and a single database. It echoes the friction that surfaced when Discord deployed the same provider for age assurance, where users questioned both the data retention and the lack of an opt-out.

That reaction is becoming the default whenever a large platform routes identity checks through a centralized document-and-selfie flow. The trust cost is real, and it lands hardest on the platforms that introduce the gate.

Identity Assurance Is Becoming a Regulatory Cost

This is not an isolated product decision. Online-safety legislation and export-control directives increasingly treat AI models as high-risk channels, the same way they treat gambling, alcohol, and adult content. As those rules land, automated age and identity assurance shifts from a nice-to-have into a standard cost of operating a consumer AI product. Anthropic is early, not unusual.

The open question is how platforms pay that cost. Routing everyone who trips a flag into manual review queues and irreversible template storage is one answer, and it is the expensive one in both engineering and trust.

Platforms running through the deepidv verification engine take a different path. Instead of leaning on heavy manual review and stored biometric templates, deepidv establishes sub-150ms cryptographic provenance over device sensors, proving a capture is live and untampered at the source. Teams currently evaluating a Persona alternative are weighing exactly this difference: structural telemetry that avoids building a permanent database of facial-geometry data.

The result is a gate that holds against fraud without asking every flagged user to hand over a credential they can never revoke.