AI rules moving into code
Regulators are turning abstract AI policy into concrete technical and legal requirements across jurisdictions. Europe says its age‑verification app is ready and the Commission has quietly clarified that agentic AI systems fall under the EU AI Act, while in the U.S. officials and lawyers are already warning about sectoral safeguards — especially for banking — and that chatbot interactions can be used as evidence in litigation ( ).
Artificial intelligence rules are moving from broad principles to product requirements, court evidence rules, and sector-by-sector controls on both sides of the Atlantic. (digital-strategy.ec.europa.eu) On April 15, European Commission President Ursula von der Leyen said the European Union’s age-verification app is “technically ready” and will be available soon for citizens using online platforms. Reuters reported that the app will work on phones and computers and ask users to upload a passport or identity card to confirm age anonymously. (ec.europa.eu; usnews.com) The same Reuters report said at least a dozen European countries, including Britain and Norway, have enacted or are considering minimum-age rules for social media, usually between 13 and 16. A senior Commission official said the app can be bypassed with a virtual private network, but described it as a barrier meant to reduce children’s exposure rather than fully police the internet. (usnews.com) In Europe, the legal frame is already in place. Regulation (EU) 2024/1689, the Artificial Intelligence Act, took effect on August 1, 2024, and the Commission says it applies a risk-based system that sets different duties for prohibited, high-risk, and general-purpose artificial intelligence. (eur-lex.europa.eu; digital-strategy.ec.europa.eu) That matters for “agentic” systems, the newer tools that do not just answer questions but can plan steps, call software tools, and take actions across other systems. The Commission’s public guidance says the law regulates artificial intelligence systems and general-purpose models by function and risk, not by whether a company markets them with a new label. (digital-strategy.ec.europa.eu; digital-strategy.ec.europa.eu) The European Artificial Intelligence Office is now the Commission unit enforcing rules for general-purpose models and supporting national authorities, and the Commission said in November 2025 that it proposed amendments to centralize more oversight of systems built on those models. (digital-strategy.ec.europa.eu) In the United States, the same shift is showing up through sector regulators and courts rather than one national artificial intelligence law. In a Reuters interview broadcast April 15, President Donald Trump said artificial intelligence could undermine confidence in banking and said there “should be” government safeguards, including a “kill switch.” (usnews.com) Reuters tied Trump’s comments to warnings from cybersecurity experts about Anthropic’s new model, Mythos, and the risks it could pose to banks running old technology stacks. Anthropic said Claude Mythos Preview would not be made generally available and declined to comment on the warnings in the Reuters report. (usnews.com) Courts are also turning chatbot use into a records problem. Crowell & Moring said Judge Jed Rakoff of the Southern District of New York ruled on February 10, 2026, and followed with a written opinion on February 17, that prompts and outputs from a public artificial intelligence tool were not protected by attorney-client privilege or the work-product doctrine in that case. (crowell.com) Reuters’ April 15 legal report said U.S. lawyers are now warning clients that chats with tools such as Claude and ChatGPT could be demanded by prosecutors or civil litigation opponents. Some firms have started writing that warning into client agreements, according to the Reuters account. (usnews.com; crowell.com) The common thread is that governments are no longer talking about artificial intelligence as a future policy debate. In Brussels, the rules are becoming apps, standards, and enforcement bodies; in Washington and New York, they are becoming banking safeguards, contract terms, and evidence fights over what users type into a chatbot. (digital-strategy.ec.europa.eu; usnews.com; usnews.com)
