Anthropic Just Changed What Claude Can Be Used For

Morning, {{first name | folks}} Today’s AI news spans new rules for high-risk AI use, agents that can work across business apps, AI-powered influence operations, interactive simulations of the physical world, and how Claude is helping astronomers map the unseen sky. Here are the five updates worth knowing today.

Anthropic has revised its Usage Policy, adding tighter restrictions on deceptive election campaigns, surveillance without consent, and assistance with weapons development. The update also introduces a rule against repeatedly subjecting Claude to needless abusive or cruel behavior, while making clear that ordinary criticism, frustration, and model testing are still allowed.

The changes come as AI systems take on more complex tasks with less human supervision. Anthropic is clarifying where it draws the line as people use Claude for increasingly sensitive and autonomous work.

Google Cloud is bringing Gemini closer to an always-on digital coworker with a new agent designed to handle complex tasks that can run for hours or even days. It can work across business apps, data systems, and coding tools, break larger jobs into smaller tasks, and delegate work to specialized agents.

The system can also switch between models, including Gemini and Claude, depending on the job. With persistent memory and enterprise controls, Google is betting that the next step for AI at work is software that can carry a task through to completion, rather than waiting for someone to prompt it at every step.

OpenAI says it disrupted influence operations linked to Russia and Iran that used ChatGPT to produce deceptive content. One posed as a research centre publishing political commentary in Latin America; another used fake journalist identities to pitch articles to online publications.

The operations used AI to draft articles, social media posts and other material designed to make their narratives appear credible. The cases show how influence campaigns can use AI to produce content at scale and disguise who is really behind it.

Odyssey has launched Odyssey-3, a world model designed to predict how environments change when objects move or actions take place. Unlike models that mainly generate images or video, it can create interactive environments that respond to what a person or AI agent does.

The company says the model can support tasks across robot arms, humanoids, autonomous driving, and drone navigation. In one test, a driving system learned to navigate real roads in India using around 20 hours of driving data. Odyssey is positioning the model as a way to train and test physical AI systems in simulated worlds before deploying them in the real one.

Anthropic has used Claude Science to help astronomers fill gaps in an ultraviolet map of the sky using decades of telescope observations. Claude coordinated multiple AI agents to gather data, resolve inconsistencies and estimate what the unobserved regions might look like.

Researchers tested those estimates against known observations, with errors of around 10% in the regions evaluated. The project shows how Claude can help scientists work through large collections of astronomical data to build a more complete picture of the sky.