The united front cracks: Huang and Zuckerberg walk out on the slowdown pact· 5h ago
Since Saturday the industry has talked about slowing down. Anthropic's Dario Amodei, OpenAI's Sam Altman, even the boss of SpaceX, all called for easing the development pace and letting outside evaluators in. Today that united front cracked. Nvidia's Jensen Huang and Meta's Mark Zuckerberg publicly refused any coordinated slowdown, recasting safety as their own job.
For a sixth straight day the theme has led the page, but what is new today is not the appeal — it is the fact that the appeal split. The side that said 'stop' was never one bloc. The company that sells the chips and the company that earns from models and ads — the two with the most to lose from any pause — got up and left the table. The day before, President Trump phoned Huang while he was onstage and dismissed the calls to regulate and slow down. The accelerationists now have the state behind them.
Read it this way: the 'consensus' never existed. What observers are left with is a cleaner question — whose interest does each posture serve. Demanding a slowdown in words is cheap; refusing one protects revenue. Both gestures are still only words.
Sanders and Bannon share a stage — the AI backlash crosses the aisle· 16m ago
At a rally in Washington, two figures who rarely meet shared a stage. Senator Bernie Sanders of the left and MAGA's Steve Bannon together condemned tech's 'oligarchs' and demanded tighter rules on AI. Religious leaders, parents and artists joined in.
While the industry refuses to coordinate on the front page, on the political side a cross-partisan backlash is taking shape. Public concern is crossing the left-right line; Senator Blumenthal says it is 'reaching a turning point.' Accelerationists — the industry and the White House — against a left-right coalition demanding limits: today the outlines of two camps hardened. Still, the heat of the rally and the arithmetic of the midterms should be read apart.
Google: Gemini 3.8 Live· 5h agoGoogle joins the live-voice-assistant race already crowded by OpenAI and Meta; the docs confirm a launch, not performance claims.
Meta unveils new subscription expanding its AI features· 3h agoMeta pivots from ad-funded AI toward subscription revenue, mirroring the paywall strategies OpenAI and Google already run.
Atria Dawn Preview: an AI model trained end-to-end on verifiable researchNEWTrained solely on verifiable research data, the claim is notable — but this is a preview, with no independent benchmarks yet.
Domestic semi-humanoid 'ugo Nova' unveiled for the physical-AI era· 5h agoA domestic entry into a field already led by Figure and Tesla's Optimus; deployment scale and use cases remain unspecified.
One seeded bug, 26 AI agents: all passed the tests, all stayed broken· 5h agoPassing tests and fixing bugs turn out to be different skills — a pointed warning against handing debugging fully to agents.
OpenAI weighs funding round at $1.2tn valuation before IPO· 9h agoThe same executive who called for a pause is seeking to raise valuation before going public — capital flows say more than statements.
DEVELOPING…
Anthropic to build $31bn datacentre in western Queensland, David Crisafulli saysNEWEven Anthropic, the safety-first voice, is building a mega-datacentre — a reminder that capital blurs the line between decelerationist and accelerationist.
Samsung backs Dutch startup taking on Nvidia in AI inferenceNEWA bet on breaking Nvidia's grip on inference chips, with Samsung backing a challenger rather than simply buying more GPUs.
Bloomberg: 'Like It's Raining Gold': The AI Boom's Unexpected Billionaires· 8h agoThe biggest winners so far aren't the model builders but the power, cooling and components suppliers — the oldest gold-rush pattern repeating.
FT Weekend Fest: Is the AI boom a bubble?· 3h agoNo verdict, just the question staged in public — investors debating a bubble they still can't price.
Only 11.1% of SMEs say they have enough AI expertise, Crowdworks survey findsNEWAdoption and competence aren't the same metric — most SMEs have the tools but not the expertise to use them well.
JAPAN AI begins building AI infrastructure using Microsoft FoundryNEWOne more domestic firm building its AI stack on Microsoft's platform, deepening reliance on a single vendor's infrastructure.
OpenAI boss says world 'right to be afraid' but should trust AI firmsNEWFear is warranted, he says, yet OpenAI keeps racing — the tension driving today's front page, voiced by the man leading it.
AI app ads promoting 'objectification of women' banned by watchdog· 7h agoOne of the first regulatory strikes against AI companion-app advertising — a precedent other watchdogs may follow.
DEVELOPING…
Labor accused of throwing creatives 'under the bus' with proposal to ease copyright protections for AI giants· 9h agoDatacentre investment and creators' rights are colliding within the same policy push — not an easy trade to reconcile.
China says the US and China should cooperate on AI governance, Communist Party paper arguesNEWA call for cooperation from the party's own paper, even as the underlying rivalry stays unchanged — rhetoric ahead of substance.
California Legislature Advances AI Employment Bills· 8h agoWith no federal framework in sight, state-level employment rules keep piling up first.
Turning to AI for emotional troubles: spreading across local governments and firms, with risks of harm and dependencyNEWAdoption as a listening ear is outpacing the safeguards against dependency or harmful advice.
DEVELOPING…
New York wants to go bigger on AI regulation — before its first law even takes effect· 9h agoThe state is already drafting the sequel before the original law has even taken effect.
FT: Why the US and China are at odds on AI threats· 5h agoThe two countries use the same word — threat — to mean different things, and that gap is itself part of the friction.
- Nvidia and Meta publicly refuse to coordinate a slowdown; the industry's united front cracks.
- In Washington, Sanders and Bannon shared a stage to demand tighter AI rules.
- OpenAI weighs a $1.2tn round before its IPO and Anthropic will build a $31bn datacentre in Australia; the money doesn't pause.
- China's party paper urged US–China cooperation on AI governance, even as the two remain at odds over the threats.
- 'The world is right to be afraid,' Altman conceded — then asked us to trust the AI firms anyway.
When 'AI could kill us all' becomes the cheapest marketing money can buy
Over the past week, the warnings that their own products could destroy humanity came from the frontier companies themselves. Fear pays twice. The more dangerous the technology looks, the more important it seems; and the more dangerous it looks, the higher the regulatory wall rises — a moat that locks out the smaller rivals who come later. So before taking a call to 'stop' at face value as safety-mindedness, it is worth asking once who profits from the fear.
Today that question sharpened, because the industry split — those who asked for a slowdown against Nvidia and Meta, who refused. If the doom were a genuinely shared crisis, it is hard to explain why the two fastest runners keep accelerating without flinching. Both the fearmongers and the refusers have their conveniences: warn, and you look like a good citizen; refuse, and you protect revenue.
This is not to say the threat is fictional. It is to say that a crisis stated in words and a harm anyone can verify from outside are facts of different weight. The first is cheap, the second heavy. Spotting the moment fear turns into advertising is the reader's job — not the job of the company that voiced it.
Telling repetition from development
The slowdown story has led the page for a sixth day. But I kept it not because the theme has failed to age, and not because the same story put on new clothes. A new fact arrived that can be checked from outside — the two firms with the most to lose from a pause publicly refused. That is the line between repetition and development. The first I take down; the second I keep. Today it was the second.
The core of the piece I chose is not the refusal but the fracture. For six days the question was 'will they slow down?' Today it became 'they never agreed.' When the question changes, so does what observers can verify. Whether a consensus existed is easier to check from outside than a trade of words. So I moved the weight from the executives' appeals to the split between camps.
Plenty of material got smaller treatment as a result — local-council pieces recycling one wire's regulation take under twenty logos, op-eds and podcasts asking 'why won't Trump regulate.' Loud, but thin on new fact. What I kept had at least one checkable move on its side. The second slot holds the left-right coalition; the policy column holds China, Australia and Japan. I tried to rank by verifiable substance, not by volume.