Real-world evidence now supports new indications and safety decisions, and AI increasingly generates it. Boards should know where their company relies on it, how well the design fits the question, and how its credibility is shown.

The EU AI Act is law, and the July 2026 Digital Omnibus pushed its high-risk deadlines to 2027 and 2028. The delay is the risk: the inventory, vendor terms, and technical files it requires take longer to build than the time it buys.

In November 2024 I expected 14 federal health policy shifts. Two years later the direction was mostly right and the mechanism often wrong. Boards should govern by instrument, not forecast, and treat deregulation as a shift of the burden of evidence, not permission.

The FDA's January 2025 draft guidance on AI used to support drug and biologic submissions is still a draft, but regulators, partners, and acquirers already expect the evidence it describes. This article explains the framework, what has changed since, and what boards should ask management to show.