Trump administration asks OpenAI to limit next model release over security concerns

Oh, what a delightful about-face. The Trump administration—so keen on deregulation that it appointed a crypto bro to oversee AI policy—has suddenly discovered the concept of 'security concerns' and decided that GPT-5.6 is a bit too spicy for public consumption. They've formally asked OpenAI to limit the next model to 'a small set of government-approved partners,' which is a euphemism for 'we don't trust the plebs with this one, chaps.'

This from the same crowd that rolled back every safety guideline they could find, that celebrated the 'AI Manhattan Project' as the pinnacle of American ingenuity, and that spent months telling us that open release is the only way to beat China. But now, faced with the actual product, the brave free-marketeers are suddenly reaching for the emergency brake. It's almost as if they've realised that AI might be a bit more dangerous than a tax cut.

The sheer, glorious irony is that OpenAI, the company that once promised to 'build safe AGI,' is now being told by the least safety-conscious administration in recent memory to pump the brakes. One can only imagine the internal meetings: 'So, wait—you want us to release it to the Pentagon and some handpicked allies, but not to, say, a researcher in Ohio?' 'Correct.' 'And this is because… it might be misused?' 'Yes.' 'But last month you said it was fine to train on copyrighted data.' 'Yes, well, times change.' How reassuring.