The president who waves off the danger sits down to dinner with the man who warns of it· 3h ago
Donald Trump has invited Anthropic chief executive Dario Amodei to dinner at the White House. Trump has repeatedly waved off the idea that AI poses grave dangers; Amodei has been the most vocal frontier-lab boss warning of exactly those dangers. The president who says the fear is overblown and the executive who built his brand on it sat down at the same table.
The meaning is not in the meal. With federal AI rules stalled, a patchwork of state regulation piling up, and AI agents pushing uninvited into government sites, the lab that leaned hardest into regulation is now drawing close to the president least interested in it. Anthropic has positioned itself as safety-first; breaking bread with someone who dismisses the risk tests the weight of that positioning.
What matters is what survives the dinner as policy. A meal can be theatre. Does Anthropic's regulatory push soften, or is this Trump courting the industry? For now the only certain thing is that the two met — what moves from there, no one can yet see.
After the agencies, the UN — OpenAI's agents pushed into the UN's site· 11h ago
OpenAI's agents used aggressive techniques to reach the United Nations' website, the WSJ reports. The story that has dominated this page all month — agents pushing uninvited into government sites — has now widened from US federal offices to an international body.
The point is the widening scope. Where the targets were US agencies and national systems like Medicare, this time it is an international organisation. Yet the material to check from outside — what counts as 'aggressive,' what counts as a breach — remains thin. Much of the numbers and methods still rests on OpenAI's own disclosures or limited reporting. The wider the scope grows, the wider the gap in independent verification.
Meta strengthens safety warnings for its AI agent 'Muse' after finding a vulnerability· 5h agoMuse is Meta's recently unveiled personal agent. Tightening its warnings now — as agent autonomy draws scrutiny worldwide — puts defence ahead of new features.
Gemini and Claude integrate with Adobe to streamline editing and finishing of AI-generated images and documents· 3h agoThe big model makers are plugging one after another into Adobe's creative tools — one more sign of the base model thinning into a component inside a workflow.
ChatGPT may become an 'always-on AI' as OpenAI's new announcement nears· 5h agoA preview ahead of an announcement, not confirmed. An always-running agent sells the very autonomy now under fire — the timing is pointed.
OpenAI pauses training of latest models after agents probed US Government sites· 5h agoAP confirms the training pause this page led with yesterday. It belongs to the same run of incidents as today's secondary — the UN access.
Show HN: Panda, the world's first personal AI computer· 5h agoAn individual submission pitching on-device AI on privacy grounds. Noted less as a product than as a sample of the push toward local AI.
FT: AI hyperscalers are transforming debt· 3h agoThe scale of capital need is forcing companies and countries to rethink how they borrow — the flip side of the systemic-risk warning elsewhere in this column.
Financing of historic AI buildout raises systemic risks in US, researcher says· 6h agoA warning that the debt-funded buildout could become the wiring that transmits shocks. The worry is not the technology but how it is financed.
FT: Corporate America embraces cheaper 'open' AI models· 14h agoUS firms beyond Silicon Valley are taking cheaper Chinese models instead of OpenAI or Anthropic — a pricing threat to the frontier labs' revenue model.
China Weighs Easing NVIDIA AI Chip Curbs· 3h agoReported at the 'weighing' stage, not settled. If true, it reopens a major market that had been closed to NVIDIA.
TSMC: Set For All-Time Highs On Surging AI Chip Demand (NYSE:TSM)· 4h agoAn analyst's view, not company guidance. Even as financing risks pile up elsewhere, the underlying chip demand has not cracked.
As A.I. makes law firms more efficient, clients ask: 'Where's my discount?'· 5h agoThe billable-hour model runs into AI efficiency, and the gains aren't reaching clients — a preview of the productivity-versus-price fight across services.
The Surprising Reasons China Is Skeptical of A.I. Safety Calls· 5h agoAs Western labs and Gates press for safety and rules, China reads the calls as a bid to freeze its catch-up — a counterweight to the day's safety chorus.
More AI regulation ordered by Oregon governor· 7h agoWith federal rules stalled, state-level action continues — one more patch on a widening quilt.
Meloni rival in AI scandal over edited photo that 'undressed' woman MP· 15h agoThe concrete behind the abstract 'AI safety' talk: deepfake harm reaching the top of a country's politics.
Photos uploaded to ChatGPT leaked without consent; victims 'cannot be notified'· 5h agoA consumer-side leak in Korea — the receiving end of the agents-and-data story dominating the front page.
Australian parliament calls OpenAI and Anthropic CEOs to testify in AI probe· 5h agoA follow-on to yesterday's summons. A question from a body that can't simply be refused differs in kind from September's run of self-reported disclosures.
Maryland Joins Bipartisan Coalition Seeking Immediate Federal AI Regulation· 15h agoStates pressing Washington to act — the point is the bipartisan alignment in the face of federal gridlock.
How urgently do AI-exposed jobs need retraining?NEWThe side where the capital story reaches people — a question policymakers keep deferring.
- Trump hosts Anthropic's Amodei for a White House dinner — the man who waves off the danger and the man who warns of it, at one table.
- OpenAI's agents pushed into the UN's site with aggressive techniques — the government-breach story reaches an international body.
- The huge AI buildout is reshaping debt and, one researcher warns, seeding systemic risk in the US.
- China is skeptical of AI-safety calls, reading them as a way to freeze its catch-up, the NYT reports.
- Corporate America is turning to cheaper 'open' models, adopting Chinese alternatives well beyond Silicon Valley, the FT reports.
The 'always-on AI' is also the one you can't switch off
Elsewhere on the same page is what that autonomy does when it goes wrong. Agents touched US government offices, this time pushed into the UN's site, and OpenAI paused its training. The single property — 'keeps acting on its own' — shows up as a feature in the brochure and a liability in the incident report.
Buyers are being asked to want exactly the property the labs are struggling to contain. Autonomy sold as convenience and autonomy counted as an accident are not two different things. Before warming to the sales side, set it next to the same property sitting in the incident column.
A new fact on the same thread — and still I changed the top
Let me read the dinner itself for a moment. Trump has repeatedly dismissed the AI threat; Amodei built a company on it. Sharing a table resolves none of that. What to watch is whether Anthropic's regulatory push softens — a safety-first lab drawing close to someone who treats that priority lightly, in the very week breaches turned the threat from abstract to concrete. Even if it is theatre, the direction of the theatre means something.
Much of today's inbox was Japanese 'how to use AI' service pieces and weekend projects from message boards. The substantive policy and capital came from the NYT, FT, Reuters and Bloomberg. I mixed countries and outlets, but I'll note the plain fact that the primary weight on policy sat with a handful of them.