OpenAI asked Washington to mandate AI safety, then signed Trump's voluntary pledge instead.

…(cont) Here's the pitch anyway: a morally binding pledge, signed at a table with over a dozen tech leaders including Amazon's Jeff Bezos and representatives from Microsoft and Palantir, is faster than a bill. It skips committee. It skips the fight a real safety statute would take to draft, mark up, and pass. Self-regulation is efficient, and efficient is safe, because—

Because nineteen days before that signing, on September 10, OpenAI's own chief global affairs officer, Chris LeHane, published a blog post asking Washington for mandatory safety requirements. Not voluntary ones. Mandatory. The company whose president sat at that table and signed a pledge to grade its own homework had, three weeks earlier, told the government the homework needed a grader.

So the same industry stood, in the same month, on both sides of the same sentence: trust us, and please don't trust us. The commitment Brockman signed carries no fine, no regulator, no enforcement mechanism the record describes — a signature and a name for the cameras, "morally binding," a phrase with no docket number attached to it.

I was going to write that the safest AI policy is the one the companies write for themselves. I'm not writing that. OpenAI already wrote the opposite, in a blog post, under its own chief global affairs officer's byline, three weeks before the signing ceremony made self-policing the official line. Campaigners and Democrats in Congress spent Tuesday calling the whole arrangement dangerous. The company that stood to gain most from being trusted had already told Washington, in writing, not to trust it.