Anthropic watermarks Claude + Gemini hits a billion users

Anthropic says Claude models launched in the European Union on or after August 2 will weave an imperceptible watermark directly into the text they generate, and will attach signed C2PA provenance metadata to supported files such as PNG and SVG. The marking is on by default everywhere Claude runs — the Claude platform API, Claude, Claude Code, Claude Cowork and Claude Tag — and Anthropic says the watermark travels with copied-and-pasted text and may survive some editing. Public detection tools are still to come, and the company cautions that a detected mark is a signal rather than proof.

Sundar Pichai said the Gemini app has passed one billion monthly users, making it the fastest product in Google's history to reach that mark and the fourteenth Google product to get there. The company had reported 950 million monthly users on its second-quarter earnings call weeks earlier. The figure counts the standalone app only, excluding people who meet Gemini inside Search, Workspace and Android — and it lands roughly two months after OpenAI's ChatGPT crossed the same threshold.

Brad Lightcap, who joined OpenAI in 2018 and served as chief financial officer before becoming chief operating officer, said he is leaving to start something new, offering only that "there are a few important new things the world will need." Separately, Chloé Bakalar — hired as head ethicist in August 2025 and reportedly the company's only dedicated ethicist — left in July and has not been replaced. OpenAI said ethics "doesn't live with one owner or team." The two follow exits by Fidji Simo, Bill Peebles and Kevin Weil.

In this week's Stratechery, Ben Thompson argues that Nvidia's new financing partnerships with Apollo, BlackRock and Blackstone — structured to mobilize $500 billion, with Nvidia backstopping up to 25 percent of residual value — move AI's capital risk out of corporate cash flow and into safety-seeking pools such as insurance floats and pension funds. He notes hyperscalers had already raised $194 billion in debt by July and Google raised $85 billion in equity. His historical comparison is Jay Cooke's 1873 railroad collapse, and his conclusion is blunt: AI had better deliver first.

Google Research reported that AMIE (Video), a version of its medical dialogue system that handles live audio and video, was rated on par with board-certified primary care physicians on history-taking, diagnostic accuracy, management and communication. The randomized study covered 100 clinical scenarios and 300 consultations with trained patient actors, scored by 20 independent physicians. AMIE rated significantly higher than the doctors at eliciting physical signs and guiding examinations. It remains research-only — every consultation used actors, not real patients.

River AI, founded in June by former xAI co-founder Igor Babuschkin, has raised $1.1 billion, led by General Catalyst and AMP PBC, with Nvidia, AMD Ventures, Y Combinator and Temasek joining. The company sells an API for reinforcement learning and fine-tuning on open-weight models, pitching agents a customer trains into their own rather than a rented model they cannot change. It claims an enterprise can finish a complex reinforcement learning run in 15 to 20 minutes with no infrastructure team.

Artists can now self-declare as an "AI Persona" in Spotify for Artists, and from mid-September the badge starts appearing on profiles, in search and on playlist track rows. Spotify will also apply a "Likely AI Persona" label itself, using a mix of human and automated review of profiles whose name and imagery appear to be photorealistic AI-generated identities. Those artists are excluded by default from editorial and algorithmic recommendations unless a listener follows them. Spotify stresses the badge is about the artist's public identity, not about how the music was made.

British Transport Police ran its first live facial recognition deployment inside a London Underground station, at Victoria on August 11, extending a trial that began at surface rail hubs in February. Cameras compare passing faces against an authorised watchlist tied to robbery, knife crime and sexual offences, plus people wanted by the courts or breaching bail, and an officer reviews any alert before approaching anyone. Images that do not match are deleted automatically and immediately. The trial runs four months, to November.