The President Signs the First Set of Federal Rules for AI

President Biden's executive order requires companies building the most powerful AI systems to share their safety test results with the government, and directs federal agencies to set standards for everything from deepfakes to drug discovery.

President Biden signed an executive order on artificial intelligence at the White House this afternoon, the most sweeping action any government has taken on the technology. An executive order is not a law — Congress has not passed one, and may not — but it directs the federal government, which is the country's largest buyer of technology, and it uses a Cold War-era law to reach into private companies.

That reach is the notable part. Any company training an AI system above a certain size must notify the government and share the results of its safety tests before the system is released. The threshold is set high enough that only a handful of systems in the world qualify today. The order also tells the Commerce Department to develop standards for labeling AI-generated content, so that a photo or a voice recording can be marked as synthetic; asks agencies to study the effect on jobs; and directs work on AI for drug development and climate research.

Biden, who reportedly became engaged with the subject after watching a Mission: Impossible film about a rogue AI, said the technology's promise and danger both demand it be governed. "We're going to see more technological change in the next ten years than we saw in the last fifty," he said.

The order is being welcomed by most of the AI companies, which helped shape it, and criticized from two directions: by those who say it does too little without a law behind it, and by those who say the reporting rule will let the government pick winners. Britain hosts an international summit on AI safety in two days. The timing is not an accident.