Superintelligence becomes policy: the US task force and the industry pledge

For years, superintelligence was debated in papers and on podcasts. In the past few weeks it became a matter of national policy in the United States. On September 29, the leaders of the largest AI companies signed a voluntary safety accord at the White House. On October 4, President Trump announced a federal Super Intelligence Force. This article sets out what each one is, and what it would take for either to make AI more reliable.

The accord

The agreement was signed by Anthropic's Dario Amodei, Google's Sundar Pichai, Meta's Mark Zuckerberg, OpenAI's Greg Brockman, Nvidia's Jensen Huang and Elon Musk. It is voluntary, with no legal penalties, and was described as morally binding. The companies committed to four layers of controls and audits:

• Internal controls: monitoring models during training and after release, including for cyber, biological and chemical threats, and for attempts by AI systems to reach systems they were not meant to.
• An internal safety team that checks those controls work and that problems get fixed.
• Independent external audits of whether the controls, monitoring and detection work as intended.
• Oversight by an independent board committee that receives the internal and external reports.

The companies also agreed to meet regularly on safety standards and best practices.

The task force

The Super Intelligence Force was announced in a post on Truth Social. It is led by the Director of National Intelligence, Jay Clayton, with the Federal Trade Commission chair Andrew Ferguson, the Pentagon's under secretary for research and engineering Emil Michael, and Office of Personnel Management director Scott Kupor. According to the Wall Street Journal, as reported by TechCrunch, it has 120 days to report on the risks and opportunities of AI, and its charter includes developing plans for responding to threats enabled by superintelligence. The President also directed the federal government to use the term super intelligence instead of artificial intelligence.

The framing puts competition first. Clayton told the Wall Street Journal that one of the biggest risks is "not being first."

The wider picture

The US is not alone in moving. The International AI Safety Report, published in February 2026 by more than 100 experts from over 30 countries, counted twelve companies that published or updated frontier AI safety frameworks in 2025. It also described an evidence dilemma: AI changes fast while evidence about its risks arrives slowly, so acting early may lock in measures that do not work, while waiting may leave people exposed. Others want a pause: a statement hosted by the Future of Life Institute in 2025 calls for a prohibition on superintelligence until there is scientific consensus it can be built safely and controllably, and strong public support.

Every layer runs on evidence

Whatever one thinks of the politics, the four layers in the accord share one dependency. Monitoring needs tests that catch real problems. Safety teams need to know what a failure looks like. External auditors need standards to audit against and people qualified to apply them. Boards need reports they can trust. None of this works without good evaluations and the expert judgment behind them.

An audit is only as good as the tests it runs and the people who read the results.

What to watch

• The task force's report, expected within 120 days, and whether it sets testing or reporting requirements.
• Whether parts of the accord move into law or regulation, as some observers expect.
• Who the external auditors will be, and what standards they will use.

We will follow each of these as it develops. The direction is already clear: claims about safe, capable AI will increasingly have to come with evidence.