Effective July 8, 2026, the revised policy covers only consumer Free, Pro, and Max accounts and leaves both the trigger for verification and the consequences of refusal unspecified.
Anthropic has updated its consumer Claude privacy policy to permit identity and biometric verification, including collection of a government-issued ID image, a photo or video of the user, facial geometry templates, and the verification result. The new rules take effect July 8, 2026 and apply to Free, Pro, and Max subscribers; Team, Enterprise, and API customers are not covered Anthropic Privacy Policy. The change, first reported by The Register, puts specific data types on the table that were not previously listed for consumer users.
The enumeration in the policy is concrete. Anthropic says it may collect an image of a government ID "and the information on it," which can include the ID number and date of birth, a photo or video of the user, a facial geometry template (a mathematical map of the face that some jurisdictions classify as biometric data), and a pass-or-fail result Anthropic Privacy Policy. The same policy update broadens the standards under which Anthropic can share user data with authorities, removing some of the previous specific-purpose guardrails.
Two questions a subscriber might reasonably ask are not answered in the text. The policy does not define what triggers a verification request beyond the stated purpose of "help[ing] keep our services safe and secure." It does not state what happens to an account that declines. Anthropic declined to comment when The Register asked about the change, leaving the practical risk of a routine interaction undefined.
The update landed one day before Anthropic released its Fable 5 and Mythos 5 models. Both are currently disabled to comply with a US government export-control order, and a coalition of more than 60 cybersecurity and technical experts has signed an open letter at freefable.org protesting that order. Anthropic has separately framed distillation attacks by foreign rivals as a model-extraction threat in a post on detecting and preventing distillation attacks. Against that backdrop, the new verification authority reads as a tool being readied rather than a routine compliance edit.
Workarounds for the existing export limits are already in circulation. Account sharing, third-country routing, and developer-side obfuscation have all been documented as ways subscribers in unsupported regions can still reach Claude models. If verification is meant to close those gaps, an undefined trigger leaves individual users guessing about who gets asked and on what signal. The clearest scope boundary in the policy is also the most basic one: the people most likely to be affected are individual subscribers, not the enterprise contracts Anthropic sells into regulated industries.
The next test is what Anthropic does with the new authority on July 8. A narrow policy that fires only on hard signals, such as a flagged account or a flagged region, is a different product from a discretionary check tied to the export-control moment. The text, as written, leaves room for both.