Claude Turns Access to AI into a KYC Verification — and It's Changing the Rules

Some Claude users are now being asked to verify their identity using a passport, driver’s license, or ID card, and then complete a selfie verification via Persona. For the chatbot, this represents a shift from a “sign up and start using” model to a “complete KYC first, then gain access to the tool” model. And the most frustrating part here isn’t the photo of the document itself, but the fact that the decision-making process lacks transparency.

Anthropic does not specify exactly under what conditions the verification is triggered. The help section mentions access to certain features and scheduled platform integrity checks. Hacker News users have noted, however, that the verification request often appears right before an account is suspended. While this doesn’t prove an automatic connection, it creates a very practical problem: the user doesn’t know whether they’re being verified to continue using the service or if their account is about to be suspended.

It’s Not Just an Age Verification

According to the source material, Persona can conduct a much broader verification process than a simple document check. The description mentions watchlists, verification of politically exposed persons, negative posts, and 269 separate checks. It also mentions modules for generating reports for the U.S. FinCEN and Canada’s FINTRAC.

In other words, this isn’t just a harmless “prove you’re 18” check. This is an infrastructure similar to that used by financial institutions for KYC and compliance. Users provide more than just a passport: their data may be processed by Persona and its subprocessors. Anthropic states that biometric data is deleted within 30 days, while other data may be stored for up to three years. According to the article, the full list of 17 subprocessors is not publicly disclosed.

This presents a particularly unpleasant trade-off. Anthropic may attribute the vetting to sanctions restrictions, age requirements, efforts to combat abuse, and regulatory pressure. All of these reasons seem plausible. But a plausible reason does not necessarily make a procedure proportionate.

Especially since the source material describes an incident involving Persona: in February 2026, 2,456 source code files for the company’s government dashboard were exposed to the public due to a configuration error. This doesn’t mean that Claude users’ credentials have already been leaked. But after an incident like this, the choice of this particular provider becomes a question that Anthropic must address in great detail.

For users, this also poses the risk of losing a work tool

Claude has long been used for more than just text experiments. People use it to write code, analyze documents, and support workflows. Therefore, a block is no longer just a loss of access to entertainment. It’s a disruption of part of their usual infrastructure.

The age verification issue exacerbates the problem. The article describes cases where adult users were blocked due to flags indicating that a child might be using their account. Unlocking the account required a separate verification through Yoti, and users were given a limited time to complete the procedure; after that, the account could be deleted.

Meanwhile, Anthropic requires users to be 18 or older, while OpenAI and Google allow users as young as 13. OpenAI has also discussed government-issued ID verification for future organizational API models, but this is not the same as requiring a passport from a regular ChatGPT Plus user. According to the source, Google Gemini does not have a similar verification process for regular users.

For Russian users, there’s an additional layer of uncertainty: Russia is on the list of countries that Anthropic does not serve. Those who currently use Claude via a VPN, a foreign phone number, or intermediaries are already in a gray area. Passport verification could make such a setup not just inconvenient, but unworkable.

What Does This Mean in Practice?

The main conclusion is not that Anthropic is necessarily acting maliciously. Rather, the company is building a system in which the service simultaneously sees the user’s queries, enforces rules of acceptable behavior, and verifies the user’s real identity. For journalists, researchers, and people who work with sensitive documents, this poses a distinct risk: their query history could potentially be linked to their passport information.

Therefore, the sensible response is not to panic or unconditionally abandon Claude, but to reduce dependence on it:

• Do not rely on a single cloud-based AI service for critical workflows;
• test local models and tools like Ollama in advance;
• for sensitive data, consider deploying the model on your own infrastructure;
• have a backup account or an alternative service ready before your primary access is restricted;
• Read the terms of data storage and processing before verification, rather than treating the request for a passport as a mere technical formality.

A powerful model may remain the best tool for code or text. But the quality of responses does not negate the cost of access. If a standard chatbot requires full identification, users must evaluate not only the model itself but also the entire data storage chain, the rules for account suspension, and the ability to leave without interrupting their work.

Source: vc.ru