The change has gone almost unnoticed. Anthropic published on April 14 a dedicated page for Claude's identity verification. Some features, security actions, or « platform integrity checks » now trigger an official ID request and a live selfie. The selected partner is called Persona, a KYC specialist in finance.

Neither law nor Washington: a trajectory already seen among crypto platforms

No legal obligation imposes this sleight of hand on Anthropic. The decision is purely internal, and it is precisely this that is causing frustration in the user community. The trajectory resembles that of crypto platforms between 2021 and 2023. Initially anonymous by design, Binance and Coinbase eventually required passports and selfies. Their revenues exploded, and regulators closely scrutinized them. In June 2023, the Paris Public Prosecutor's Office opened a preliminary investigation against Binance for aggravated money laundering. The platform did not impose KYC on French users before August 2021. A year later, the UK FCA imposed £3.5 million on Coinbase's payment subsidiary. The first sanction by an European regulator against a cryptocurrency-related entity.

The company aims for a public listing in autumn 2026. Its annualized revenue approaches $30 billion, with a valuation exceeding $380 billion after the February Series G funding round. Before going public, a company must be able to prove that it knows its customers. OpenAI followed the same path a few months earlier. The rival deployed AI age verification in December 2025, requiring ID for some users mistakenly categorized as minors.

An AI that is already monitoring its users, its own code leak revealed it

The announcement falls weeks after a leak that left traces. On March 31, a .map file accidentally divulged the 512,000 lines of source code for Claude Code, published via an npm package. Developers immediately dissected the content. One discovery caught attention. The file userPromptKeywords.ts contains a regular expression that detects curses, insults, and expressions of frustration in messages. The « wtf », « ffs », « this sucks » and equivalent French expressions are identified and then recorded as negative sentiment signals.

The creator of Claude Code, Boris Cherny, confirmed the mechanism on X. Internally, the feedback lands on a dashboard humorously named « f***s chart ». The company also adds a sentiment classification, reserved for employees. Combined with the new requirement of passport and selfie, the picture becomes dizzying. Anthropic knows what you say when you complain. The company will soon know what you look like, and will have a copy of your ID document with its partner Persona. Ah yes, don't forget the possible profiling simply through the astronomical amount of chats and the memory function.

The risks are not theoretical. In October 2025, a breach at Discord exposed approximately 70,000 identity documents submitted for platform age verification. Persona remains stronger than average: no leaks have been documented so far, but no subcontractor is infallible. For a company that had built its image on rejecting Pentagon contracts and protecting privacy, the turn has a bitter taste.

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source: Support Anthropic