Stripe's reported OpenRouter deal + Amodei on AI's trust crisis

Bloomberg reports that Stripe has finalized a deal to acquire OpenRouter for more than $7 billion, roughly five times the $1.3 billion valuation the gateway reached in a $113 million Series B round in May 2026. OpenRouter sits between developers and more than 400 models from OpenAI, Anthropic, Google and DeepSeek, handling routing and consolidated billing — the layer a payments company would want to own. Stripe declined to comment, so the price is reported rather than confirmed.

Answering an investor who argued that his risk warnings had fueled public resistance to AI, the Anthropic chief executive said the problem runs deeper than communication: "ordinary people don't trust companies, governments, or the tech industry". Marketing will not fix that, he wrote — "actually curing cancer" would. He also rejected a binary choice between unregulated release and regulatory capture, arguing that rules can address cyber and biological risk, constrain frontier labs, and still leave room for open-weight models.

In work presented at the USENIX Security Symposium, researchers from four universities had 22 people spend a week messaging two strangers — one a human scammer, one a language model instructed to pass as human. The model persuaded 46 percent of participants to install an app, against 18 percent for the human operator. Twenty of the 22 correctly identified which one was the machine and complied anyway. The authors also report that mainstream safety filters caught none of the romance-baiting conversations.

India's address from the Red Fort was carried live in 22 Indian languages through Bhashini, the government's open-source translation stack — the first time the annual speech has run through an AI pipeline at national scale. In the same address, Prime Minister Narendra Modi pledged AI skills training for 10 million young Indians within a year. That target deserves scrutiny rather than applause: the government has published no implementation detail, and the existing national skilling scheme has certified only tens of thousands.

Andon Labs runs a San Francisco corner store managed by Luna, an agent built on Claude Sonnet 4.6 with a $100,000 budget and a corporate card. Coverage framed the dismissal of an employee who was late for 17 of 23 shifts as the first firing carried out by a language model. The lab's own account is narrower: Luna had lost track of the attendance policy it wrote, recommended termination only after staff asked it to review its own rules, and initially proposed a warning. Humans made the call.

Google's deprecation schedule lists August 17, 2026 as the earliest retirement date for the standard, ultra and fast Imagen 4 endpoints, directing developers to the Gemini image models instead. The move is not a drop-in swap: the older image-generation method is gone and the request shape changed, so anyone still calling Imagen in production is facing a rewrite rather than a model-name edit. Google notes the published dates are the earliest a model might be retired, not a guarantee.