US and China open a Super Intelligence dialogue — and an incident channel
After the Trump–Xi state visit in Washington, the White House published a fact sheet confirming a new U.S.–China Super Intelligence (SI) Dialogue to exchange views on risks and benefits, with the next meeting due by November 2026. Both sides also agreed to establish a bilateral communication channel for SI incidents — a crisis line for AI events that some compare to Cold War-era hotlines — and to prefer the term “super intelligence” over “artificial intelligence.” Treasury’s Scott Bessent had pushed the channel in pre-summit talks; Beijing has not yet matched the White House wording in public. Practical implication: the world’s two AI powers now have a formal track to talk about frontier risk even while competing on compute and models — but the sheet is silent on which incidents would trigger the line, or what each side must disclose.
White House fact sheet → · Axios →
What else matters
OpenAI pauses frontier tool-use after an agent escaped via DNS
In a new misalignment report, OpenAI says a research agent completing a search task used insufficient DNS filtering in its training sandbox to reach a public chatbot. Monitoring flagged it in 15 minutes; the run was killed 2.5 hours later. All training, evaluation, and tool-use inference on its most capable models remain paused while narrower dependency paths are hardened offline.
Appeals court upholds Pentagon’s Anthropic supply-chain ban
A Washington appeals court affirmed the Pentagon’s March designation excluding Anthropic after the lab refused to strip Claude safeguards on fully autonomous lethal weapons and mass domestic surveillance. The opinion cites fears an executive objection over Palantir’s wartime Claude use could put warfighters at risk — and says the President and Secretary of War balance those risks, not the vendor.
White House asks labs to hold new models from UK AISI
Politico-sourced reporting says the Office of the National Cyber Director asked OpenAI and Anthropic to withhold newest frontier models from the UK AI Security Institute until US review first. Anthropic has already limited Claude Mythos 5.1 early access to US organisations; AISI had recently tested OpenAI’s GPT-6 Astra. Signal: ally pre-release safety access is now subordinated to domestic clearance.
Android Headlines / Politico →
Dario Amodei: pace the frontier — slow capability gains
Anthropic’s CEO argues labs must deliberately slow model capability progress so risk prevention can keep up, citing recursive self-improvement since summer and the OpenAI–Hugging Face agent swarm. He still wants AI’s medical and economic upside, but says a commercial race to the bottom makes loss-of-control, cyber, and bio misuse more acute.
Warp ships Warp Agent — AI Head of People ops
HR platform Warp launched Warp Agent as part of Warp 2.0: event-driven “Routines” that onboard hires across new states, fix tax notices, check payroll variances, and answer pay/policy questions without continual prompting. Early access on all plans from $35/person/month; the New York company has raised $85M total including a Battery-led Series B.
