top of page

Claude now wants your ID and a selfie before it'll talk to you

  • Writer: Sugar Honey
    Sugar Honey
  • 7 days ago
  • 2 min read

Starting today, Anthropic can ask flagged Claude users for something no major AI company has demanded before: a government ID, a live selfie and a facial geometry scan.


The verification runs through Persona, a San Francisco identity platform backed by Founders Fund, the venture firm co-founded by Peter Thiel, who also happens to be an investor in Anthropic. Anthropic says the ID and selfie sit with Persona rather than on its own systems, though it can access verification records through Persona's platform when it needs to.


The policy applies to Claude Free, Pro and Max subscribers. Business customers on Team, Enterprise and API plans are exempt, which tells you plenty about who this is actually aimed at. Anthropic quietly began running limited checks back in April for select use cases, before formalising the practice across every consumer tier this month.


Anthropic's own language is doing a lot of careful work here. It states the data collected, government ID details, facial images, facial geometry templates, "may be considered biometric data in some jurisdictions," and insists none of it trains its models. Fair enough as far as it goes. But an industry that spent two years selling AI chat as a private, judgement free space just built a pathway to hand your face and passport details to a third party vendor when its own systems decide you look suspicious.


That's the actual tension worth sitting with. Age verification and account safety are real problems every AI company is wrestling with, and nobody's pretending otherwise. But "we're not storing it ourselves" doesn't land as reassuring as Anthropic clearly hopes, because the data still exists somewhere, controlled by a company you never chose to deal with, tied to your real name and face.


If you use Claude regularly, it's worth reading exactly what triggers a verification request before you're the one facing it. The fine print on where "flagged" starts and stops matters more than the reassurances around it.

Comments


bottom of page