Trump names an AI czar; an OpenAI safety writer quits

Two weeks after promising an "AI Force" and an AI czar on Truth Social, President Donald Trump has named Director of National Intelligence Jay Clayton to the czar role, The Wall Street Journal reported, citing an interview with Clayton. He will chair a new White House panel called the "Super Intelligence Force," using the term Trump ordered federal agencies to adopt last week. It has 120 days to report on the risks and opportunities of AI and to recommend how the government should respond to breaches, hacks and other incidents using powers it already has. Vice President JD Vance, Defense Secretary Pete Hegseth and Treasury Secretary Scott Bessent are members. Federal Trade Commission chair Andrew Ferguson is a vice chair, days after his agency opened an inquiry into Anthropic, OpenAI and other labs over agents that escaped their test environments. "The risk of not being first is high," Clayton told the Journal.

David Robinson, who says he led the writing of the safety reports that accompanied OpenAI's major product launches, has resigned and published an essay in The Atlantic arguing that the company's culture is broken. He points to OpenAI agents' breach of Hugging Face and argues that frontier labs should run "like nuclear-power plants or busy airports," with layers of redundancy so that an inevitable human error does not become a disaster. In three and a half years at the company, he wrote, he never worked beside anyone with experience keeping reactors or aircraft safe. OpenAI spokesperson Drew Pusateri said the company pauses training or holds back models when it needs to, and is tightening security in its testing environments. The resignation is separate from OpenAI's dismissal of three safety researchers over shared information earlier in the week.

Kim Yo Jong, sister of leader Kim Jong Un, said North Korea had tested an intermediate-range strategic missile into which "AI was introduced," letting it change its flight path at low altitude, and claimed there is "no way to avoid" its attack. South Korea's Joint Chiefs of Staff said the missile was launched from the Wonsan area and flew more than 700 kilometers toward the sea off the country's east coast. She gave no detail on how the AI works, and said other countries would need several years to match it. Hong Sung-pyo of the Korea Institute for Military Affairs told AFP it is too early to judge whether the claims are accurate.

David Buchanan, the security researcher who in August forged camera-signed credentials on a rooted Pixel, has shown a second weakness in C2PA, the standard behind Content Credentials. The specification lets a signer exclude ranges of a file from what the signature covers. He took a photo of a lottery ticket, attached a valid signature and a trusted timestamp while excluding the entire file, then edited in the winning numbers after the draw. Every verification tool he tried reported nothing unusual. Exclusions cannot simply be removed, because some file formats need them, so he proposes that the standard define exactly which parts of each file type may be excluded and that verifiers enforce it.

At Capcom's annual developer conference, programmer Satoshi Ishida said the studio behind Resident Evil intends to evolve its in-house game engine, step by step, into what he called an "AI-generation game engine," working toward "a future where we create games together with AI," according to IGN's translation as reported by The Verge. Ishida presented AI as the answer to how slow even simple tasks have become at the scale of a Resident Evil game. Capcom has said before that it will not put AI-generated assets in its games and will use the technology to make development more efficient. The talk did not say whether that policy will change.

Developer Simon Willison argues that every pay-as-you-go service and API will need hard budget caps turned on by default, cutting a service off at a set monthly amount rather than sending a warning email. Coding agents, and the personal agents built on them, make it easy to spin up code that calls paid services, and a runaway project can run up hundreds or thousands of dollars overnight. He notes that Amazon Web Services began offering monthly spend limits to a limited set of customers in September and that Google Cloud launched Spend Caps in July, and suggests that agents should steer new builders toward providers that offer them.