Trump administration considering safety review for new AI models after Mythos

Splendid news from the White House, where the Office of the National Cyber Director has finally — *finally* — held meetings about the security concerns posed by advanced AI models like Anthropic's Mythos Preview. The solution? A framework that would require the Pentagon to lead safety testing for *government AI deployments*. Because if there's one institution known for its nimble, transparent, and error-free handling of complex technology, it's the Department of Defense. I'm sure the same folks who brought us the F-35's software woes will do a bang-up job ensuring our AI doesn't go rogue. 🫡

This, of course, comes hot on the heels of the administration's earlier hints at a working group to 'examine potential oversight procedures'1 — a phrase so bland it could have been generated by a particularly uninspired language model. So now we have the Pentagon as our digital nanny, prodding at black-box models with all the subtlety of a cruise missile. The fact that Mythos was deemed too dangerous for public release by Anthropic itself seems to have registered, but the response is more 'lock the stable door after the horse has bolted' than any genuine precaution.

The irony is thick enough to spread on toast: the same administration that spent years deregulating everything from emissions to banking now wants the military to certify AI safety. I'm sure there's a perfectly rational explanation — something about 'cutting red tape' while simultaneously creating a new bureaucratic nightmare. Perhaps the real threat isn't Mythos but the meeting minutes from these sessions, which will likely be classified before anyone can point out the contradictions. At least the gallows humour is writing itself.