Governments are spending billions to define their AI independence. HAI Denning Director James Landay explains that the commercial market rushing to meet that demand offers real benefits, but may also provide options that deepen the dependencies it promises to dissolve.

As more people turn to chatbots for mental health support, AI developers have come to rely on expert ratings to validate model safety. A new study finds that when psychiatrists rate the same AI responses, their scores diverge in ways that reflect genuinely incompatible clinical frameworks, leaving developers without a reliable method for improving model safety.

From helping to build the largest DNA language model ever trained to co-creating an AI research assistant used by 15,000 scientists, Stanford HAI faculty are at the frontier of a fundamental shift in how science gets done. The barriers to scientific discovery have always been the same: too little time, too few resources, too much data to process. AI is changing that.