Opinion | Trump Is Finally Facing Reality on A.I.

Source Domain: www.nytimes.com

President Trump has spent most of his second term resisting calls to impose federal oversight over artificial intelligence. But in recent weeks, his administration has shown signs of softening its hands-off approach.

Nearly two weeks ago, the president was set to sign an executive order at an event featuring American tech leaders, which would have asked A.I. companies to send new models to the government for review before releasing them to the public. The order was scrapped hours before it was to be signed, with Mr. Trump voicing concerns it would hamstring the private sector. Now, with less fanfare, he’s signed a similar executive order that institutes a shorter review process.

Anti-regulation forces say the order goes too far. Pro-regulation groups say it doesn’t go far enough. Having spent much of my career defending critical systems in government and industry, I welcome the move as recognition of an urgent new reality: The most powerful A.I. capabilities are becoming too consequential for national security to be released without meaningful coordination between the companies building them and the government responsible for protecting the country. It is, however, only a first step to building a stronger federal strategy for safeguarding Americans from threats posed by A.I.

In recent weeks, two leading American A.I. companies offered limited access to advanced cybercapabilities. Anthropic provided select entities early access to Claude Mythos Preview, which the company says has identified thousands of previously unknown vulnerabilities in critical software. OpenAI gave restricted early access to GPT-5.5-Cyber, providing vetted entities greater ability to conduct authorized security work under additional safeguards.

Both decisions reflect a kind of self-restraint rare in an industry where competitive pressure generally favors broader and faster deployment. They are a signal of how seriously these companies judge the risks of what they have built.

Those risks are…

Source