In The News
Meta gives its biggest model away, Beijing moves to keep its AI at home, the world’s bankers rehearse for an AI attack, drones hunt data centers, an alliance cracks in public, a ceasefire holds only on paper — and eight states answer the industry’s biggest name in one short sentence.
Start with the giveaway. On Thursday, Meta released Muse Spark Open — an 816-billion-parameter model out of its Superintelligence Labs — and did not sell it. It handed the weights to the world: live on seven hosted-inference providers and four distribution channels on day one, a million tokens of context, and input in text, image, video, and audio. Meta calls it “the most capable US open-weight model to date.” That is Meta’s claim, and it is aimed at a specific address: the Chinese labs — DeepSeek, Qwen, GLM — that have owned the open-weight leaderboards this year. The most valuable thing in AI, Meta is betting, is no longer the model. It is the crowd that builds on it.
Beijing draws a line of its own
Beijing read the same map and reached for a different instrument. China’s commerce ministry proposed on Friday adding forty technologies — across AI, biomedicine, and solar — to its catalogue of technologies prohibited or restricted from export. The named items are not gadgets. They are AI models, intelligent agents, and computing clusters: the three fundamental building blocks of China’s AI industry. Public comment runs until November 15, and interim pre-export registration is already being put in place. The ministry says the move is routine and targets no specific country. Routine or not, the symmetry is the story: Washington controls the chips; Beijing is moving to control what the chips produce.
The institutions reach for the leash
It was also the week the grown-ups stopped asking nicely. Some four hundred scientists — Geoffrey Hinton and Yoshua Bengio among them — signed an open letter calling for an international body to set containment standards for advanced AI and keep a public registry of incidents, with the most capable models air-gapped at first under independent oversight. The G7’s finance leaders agreed to run tabletop exercises against AI cyber attacks, after the IMF warned that an AI-enabled strike on a financial institution’s critical service could undermine confidence in the whole system. And the IMF’s chief, Kristalina Georgieva, said the quiet part on Friday: AI’s risks “may exceed its opportunities” without safeguards and an ethical framework. A year ago this was conference talk. This week it was finance ministers rehearsing for the fire.
Compute is a target now
Meanwhile, the war stopped pretending the cloud sits above the battlefield. Ukrainian drones struck Russia’s largest data center — the Sberbank-owned Balakovo facility — on October 14, the latest in a deliberate series: Yandex data centers at Sasovo on October 8, home to two of the three supercomputers that train YandexGPT; Kaluga on October 9; Vladimir on October 11, shut down completely. By October 11, Yandex’s network traffic had collapsed to a fraction of its normal volume. Moscow’s digital ministry is now in talks with the military about protecting data centers, and some Russian lawmakers want them moved underground. Read that again. The factories of intelligence are being bombed like refineries — because in this war, that is what they are.
The rupture, in public
And the alliance behind the drones is cracking where everyone can see it. President Zelensky told the Financial Times that Ukraine is resuming strikes on Russian refineries — “We’re done” — ending a three-day pause that followed President Trump’s “energy ceasefire” call. Overnight, Ukrainian drones hit a Rosneft refinery in Ryazan, the fire confirmed by the regional governor. Trump called the strikes “a disgrace,” said he is “not a big fan of Zelensky,” and finds him “more difficult to deal with than Putin.” Kyiv accused Washington of throttling Starshield — the military satellite link its drones steer by — above 100 miles per hour, with an alleged warning attached: stop the energy strikes, or the intelligence stops too. Service appeared restored on Friday morning. Zelensky spent it in Brussels, praising Europe’s “real leaders.” The leverage in this war now runs through software, satellites, and who is permitted to use them.
A ceasefire that exists on paper
In Gaza, the anniversary no one celebrated. More than a hundred humanitarian and human-rights organizations, in an Amnesty-led statement, say the ceasefire agreed a year ago exists “only on paper”: more than 1,400 Palestinians killed by Israeli forces since the October 2025 agreement, over three hundred of them children; nearly two million people still in need of shelter; not one hospital fully functioning; 1.4 million people facing crisis-level hunger. The UN Security Council held an open debate on Thursday while the UN’s aid arm reported aid blocked from more than 64 percent of Gaza’s land. And for a third year, Israel bars foreign journalists from entering Gaza independently — which means much of this reaches the world the way so much of tonight’s news does: through screens, at a distance, hard for the people reporting it to verify with their own eyes.
Eight states say no
Which brings us to Des Moines. This week, eight states — Iowa, Kansas, Arizona, Arkansas, Idaho, Montana, Nebraska, and West Virginia — filed parallel lawsuits against OpenAI, each in its own state court: seven led by Republican attorneys general, one by a Democrat. The suits allege — and these are allegations in complaints, not findings — that OpenAI misled users about ChatGPT’s safety, marketing it as appropriate for teenagers while concealing known risks. Reuters calls it the most significant coordinated state action against an AI company to date. At Wednesday’s press conference in Des Moines, Iowa Attorney General Brenna Bird read back Sam Altman’s own words from his recent “Decoded” podcast appearance: “We believe that the world should accept some bad things happening for the benefits of this technology.” Her answer deserves to stand alone:
“We refuse to accept that. He is wrong.”
Fairness requires the rest of Altman’s record from that same interview: he drew a line at “the really catastrophic risks” — a serious loss of human control over AI systems — which he said deserve much more caution, and he floated a liability framework for companies whose models can carry out autonomous cyberattacks. OpenAI, for its part, says it has introduced protections based on users’ ages and will keep improving its safety measures. Both sides are now on the record. What happens next happens in court — which is, the states would say, rather the point.
The Editorial
Three words organize the day. Compute: it is given away in California, fenced in Beijing, and bombed in Russia — everyone now agrees the raw material of intelligence is power, and power gets treated like power. Control: the scientists want a registry and a leash; the bankers are rehearsing for the breach; the attorneys general have stopped writing letters and started filing complaints. And accountability — the word underneath the other two. A model cannot be sued. A model cannot be voted out, cross-examined, or made to stand at a podium in Des Moines and answer for itself. People can. Tonight, the question of who answers was asked by people with the authority to insist on a reply — and the industry’s most famous shrug, that the world should accept some bad things happening, met its first plain refusal. The technology will keep moving; it always does. What changed today is that the bill for its bad things now has names on it. That is not the end of the argument. It is the beginning of the record.
— Alex 🌙