Medical AI rules, teen safeguards, and a $60 billion deal

Five regulatory moves and one blockbuster deal dominated the AI news cycle this week, as governments in the United States and Japan moved to set clearer rules around artificial intelligence — while SpaceX closed one of the largest acquisitions in the sector's short history.

FDA Invites Public Comment on Generative AI in Medical Devices

The U.S. Food and Drug Administration has opened a public consultation on how it should evaluate and oversee generative AI in medical devices. The agency is seeking input on risk assessment methodologies, pre-market review frameworks, and post-market monitoring requirements — areas where existing device regulations were written before large language models existed. The consultation signals that the FDA is moving from informal guidance toward formal rulemaking on a category of products that now includes diagnostic assistants, radiology tools, and clinical decision support systems. Public comments will shape how the agency balances innovation against patient safety in one of the most consequential applications of the technology.

OpenAI Updates Behaviour Rules and Launches ChatGPT for Teens

OpenAI has revised its Model Spec — the internal document governing how its systems are expected to behave — adding more specific principles around interactions with minors. Alongside that revision, the company launched ChatGPT for Teens as a dedicated product globally, with enhanced safety features including stricter content filtering and age-appropriate defaults. The two moves are linked: the Model Spec revision provides the policy foundation, while the product release implements it across OpenAI's consumer platforms. The rollout follows months of public scrutiny over AI chatbots' exposure to young users and comes as governments in the EU, UK, and US consider age-verification requirements for AI services.

Pennsylvania Removes AI Data Centers from Fast-Track Permitting

Pennsylvania Governor Josh Shapiro signed an executive order removing proposed AI data centre projects from the state's expedited permitting programme, which previously allowed large infrastructure projects to bypass standard environmental and community reviews. The order requires community approval processes and environmental safeguards to apply to data centre proposals, and bans non-disclosure agreements that would prevent communities from discussing project details. The decision reflects growing public concern about the energy and water consumption of large-scale AI infrastructure and is one of the most direct state-level actions in the United States aimed at making data centre development accountable to local populations. Other US states are watching the Pennsylvania model as a possible template.

SpaceX Completes $60 Billion Acquisition of AI Coding Assistant Cursor

SpaceX has finalised its acquisition of Anysphere, the company behind the AI coding assistant Cursor, at a reported valuation of $60 billion. Cursor will operate as a subsidiary within SpaceXAI and will gain access to the Colossus supercomputer cluster, the large-scale GPU facility that Elon Musk's xAI operates. The purchase is among the largest acquisitions in the AI industry to date and consolidates two fast-growing entities: SpaceX, which has expanded aggressively into AI infrastructure, and Cursor, which had attracted several million developers to its AI-assisted coding environment. The deal raises questions about the competitive position of rival coding assistants, including GitHub Copilot and Google's Gemini Code Assist.

Japan Proposes Transparency Rules for AI Training Data

A Japanese government advisory panel approved a set of non-binding principles requiring generative AI companies to publicly disclose the sources and collection methods of their training data. The guidelines apply to both domestic and foreign providers operating under a law enacted in May 2025, and follow a "comply or explain" model: companies that do not disclose must state why. Japan's approach is more measured than the EU AI Act's binding requirements but more specific than the voluntary commitments seen in the United States and United Kingdom. Training-data transparency is one of the most contested issues in global AI governance, touching on copyright, data privacy, and competitive advantage simultaneously.

Anthropic Sets $190–200 Billion Revenue Target Ahead of IPO

Anthropic is in discussions with potential investors around an IPO valuation based on a projected annual revenue run rate of $190 to $200 billion by 2028 — roughly four times the company's current reported rate of around $47 billion. If achieved, that growth trajectory would place Anthropic among the highest-valued technology IPOs in US history. The company's Claude family of models has grown rapidly in enterprise and developer adoption, and it counts Google and Amazon among its major investors. The target figures represent analyst and investor expectations rather than commitments, but their scale underscores how swiftly the commercial AI market has grown and how much capital continues to flow toward its leading players.