Jay Clayton will run Trump's Super Intelligence Force
Washington gave its AI rebrand a name and a clock this Sunday, SoftBank's founder asked the world to take superintelligence off autopilot, and Google published a fix for agents that quietly memorize the tests they are graded on.
Donald Trump named director of national intelligence Jay Clayton as the White House's AI czar on Sunday, putting him in charge of a "Super Intelligence Force" tasked with reporting on AI risks and opportunities within 120 days. Clayton keeps his day job running the intelligence community — a dual-hat that officials described as effectively making him the president's AI decision-maker too. Trump's announcement says the force will coordinate federal engagement with consumers, public interest groups, religious organizations, critical infrastructure providers, and "super intelligence companies." It is the first concrete delivery on the rebrand begun by Executive Order 14434, signed September 29, which orders federal agencies to say "Super Intelligence" instead of "AI" and to stop acknowledging the old term altogether — though the 120-day clock comes from reporting on the task force's terms, not the order itself, whose only deadline is 60 days for a working legislative definition. We flagged the pick before it was official — Trump likely to name Jay Clayton as White House AI czar — and the earlier promise it grows out of — Trump creates an 'AI Force' and says a czar will run it. The rebrand now has a name, a deadline and one man holding two of the most sensitive jobs in Washington; the real test is whether a 120-day report says anything the last few haven't.
SoftBank founder Masayoshi Son says superintelligence in the wrong hands could turn "super dangerous" — and that countries no longer have the luxury of fighting each other while it arrives. Speaking Sunday on the sidelines of the Science and Technology in Society Forum in Kyoto — the same stage that produced the US-led "Kyoto Vision" statement on AI for science, backed by a group of countries that did not include China — Son argued that AI's exponential capability climb leaves nations no choice but to build trust and steer the technology together. The call for multilateral restraint lands differently coming from the man whose fund has poured nearly $65 billion into OpenAI, and he paired it with his usual ambition: AI and machines will account for at least 20% of global GDP, or $46 trillion, by 2040, he said. "We don't have any more luxury of having human fight against human." The direct quotes come from Bloomberg's exclusive; no second outlet independently obtained them, so treat the wording as one reporter's record of the remarks.
Google Cloud AI Research published a method that keeps self-improving agents from memorizing the tests they are graded on — and beat four rival approaches precisely where they fail. When an agent rewrites its own harness — the prompts, tools and workflows wrapped around a frozen model — against a fixed set of tasks, scores on those tasks climb while gains on new ones evaporate. The fix, called RRSI, caps how many edits each round may bundle with a budget that shrinks over time, while a critic discards proposals that hardcode benchmark-specific tricks and a pruner drops changes too small or expensive to justify. Across eight benchmarks with Claude Opus 4.8 held constant, it gained up to 14.1 points on the training tasks and up to 4.7 on five it had never seen, while running on about 30% fewer tokens — and it was the only method that stayed above baseline on unseen tasks, where two competitors ended up below it. A harness tuned on Gemini 3.5 Flash even lifted the much weaker Gemini 3.1 Flash Lite from 11.2 to 14.6. This is a concrete answer to the memorization worry sitting behind this week's Hinton, Bengio among 22 authors warning of an intelligence explosion: the code is public, so the guardrail is adoptable, not just argued.
What to watch: whether Clayton's 120-day report arrives before the executive order's own 60-day definition deadline — and whether "Super Intelligence" survives contact with the federal rulebook.
Can one person sensibly run the intelligence community and the government's AI portfolio at the same time? Tell us in the comments.
Sources: RRSI code (GitHub)
