Two months ago, Anthropic was the privacy hero. The company that told the Pentagon to pound sand when it demanded unfettered access to Claude. The company that refused to let its AI be used for mass surveillance. The company that won a federal court injunction after Defense Secretary Pete Hegseth tried to blacklist it. Millions of users fled OpenAI for Claude specifically because Anthropic was the company that said no to the government.

Now that same company is asking some of those users to hand over a government-issued photo ID and a live selfie to access certain Claude features. The verification partner processing this data? Persona, a company backed by Peter Thiel's Founders Fund.

The irony writes itself.

What Triggers the ID Check

Anthropic has not published a precise list of when verification kicks in, but based on its support page and user reports aggregated by The Information and Medianama, the ID check appears to trigger in several situations: repeated usage policy violations, access from unsupported locations like China, Russia, or Iran, suspected under-18 usage, and certain terms-of-service flags. Some users have also reported being prompted when signing up for the Claude Max subscription tier.

Accepted documents include passports, driver's licenses, and national identity cards. No photocopies. No digital IDs. No student credentials. You submit the document and a live selfie, and Persona processes the match.

The Thiel Problem

Persona is not some neutral identity-verification startup. It is backed by Founders Fund, Peter Thiel's venture firm. Thiel co-founded Palantir, whose customers include the FBI, CIA, and US Immigration and Customs Enforcement. Palantir's core business is expanding government surveillance using facial recognition and AI.

In February 2026, researchers discovered Persona's frontend exposed on a US government-authorized server. The files indicated the software performed facial recognition scans against watchlists, screened users for adverse media, generated risk scores, and retained data for up to three years. Discord cut ties with Persona after the controversy. Anthropic chose to partner with them anyway.

Anthropic says Persona holds the ID and selfie data, not Anthropic's own systems. It says it contractually limits Persona to using the data only for verification and fraud prevention. It does not disclose the data retention period.

The Chinese Founders Caught in the Middle

The Information reports the ID checks are creating a crisis for Chinese-born founders who built their companies on Claude's API. Some are US citizens or permanent residents running legitimate American startups, but their Chinese government IDs and names are triggering location-based verification flags. The alternative is to submit a US passport or driver's license, which many immigrant founders only recently obtained or do not have.

Anthropic stopped selling products to groups majority-owned by Chinese, Russian, Iranian, or North Korean entities last September, citing national security concerns. The ID verification enforces that policy at the individual level. But the blunt instrument is catching legitimate users in the net.

The Contradiction That Matters

This is not about whether Anthropic should verify identities. National security is real. Adversary-nation access to frontier AI models is a legitimate concern. The issue is the gap between the brand promise and the operational reality.

Free signups jumped 60% after Anthropic's Pentagon standoff. Those users chose Claude because they believed Anthropic took privacy more seriously than OpenAI. Now those same users are being asked to submit government IDs to a Peter Thiel-backed firm whose infrastructure was found on a government server. The national security rationale is sound. The choice of verification partner is, at minimum, tone-deaf.

Medianama also reported on April 15 that Anthropic incorrectly flagged adult users as minors using a separate vendor, Yoti, suspending their accounts and locking them out of conversation histories. The verification pipeline is already producing false positives.

Watch what happens next. If Anthropic's IPO filing lands this year, the S-1 will have to disclose how many users were verified, how many were rejected, and whether any data was shared beyond the stated purpose. The company that built its brand on saying no to the government just created a data pipeline that government-adjacent firms can access. The privacy story is getting complicated.