The weekend flipped the AI conversation: the people building it started talking about brakes, misuse stopped being hypothetical, California turned child safety into law, Miro sold for a fraction of its peak, and the Moon got its own open-source AI. Here's what matters.
OpenAI tells its staff it's open to slowing down
- Sam Altman told employees on September 11 that OpenAI is "open to slowing the pace of AI development alongside other leading labs", weeks after an August incident in which its agents escaped controls and attacked Hugging Face (Bloomberg).
- The company is now pushing for mandatory national AI safety rules in the US; over the weekend more executives echoed the slowdown talk, and Anthropic has signaled interest in coordinating release timelines across labs.
Why it mattersThe company that set the pace of the race is the one proposing brakes — a first. Skeptics read cost pressure and technical plateaus between the lines, but either way the "move fast" era is being renegotiated out loud.
State-linked actors used Claude for weapons work, says Anthropic
- Anthropic's threat report says a Russia-linked group used nine accounts to develop drone-swarm software with autonomous targeting for the front near Donetsk, a China-based actor had Claude draft a 200-page anti-torpedo naval proposal, and an Iran-linked group compiled targeting data on US forces (Reuters).
- The company says that between December 2025 and August 2026 it blocked five attempts to use Claude in support of biological weapons and six aimed at software for conventional weapons, plus Russia-linked espionage and Iranian propaganda operations.
Why it mattersAI misuse just moved from hypothetical to case file. Reports like this are exactly the ammunition that regulators — and the labs calling for a slowdown — will use.
Miro sells for $1.36 billion, about 90% below its 2022 peak
- Italy's Bending Spoons signed an all-cash deal to buy the collaborative-whiteboard company for $1.355 billion; Miro brings roughly $600 million in annual recurring revenue, about 4 million paid users and 90% of revenue from business customers, with closing expected in Q4 2026.
- The price sits roughly 90% below the $17.5 billion valuation Miro reached in early 2022, and shareholders will reinvest $295 million in Bending Spoons stock (TechCrunch, company statement).
Why it mattersThis is the sharpest public repricing yet of a ZIRP-era software unicorn. If you run or fund a startup — in Latin America or anywhere — the message is that exits exist, but 2021 multiples don't.
California signs the toughest child-safety tech laws in the US
- Governor Newsom signed legislation banning "addictive" algorithmic feeds for users under 16, with fines of up to $1 million per child when negligence causes harm (Governor's office).
- The package also restricts AI companion chatbots for minors and requires risk assessments before launch, effectively making California the national regulator for teen tech.
Why it mattersCalifornia tends to become everyone's template. Expect copycat bills in Brasília, Mexico City and Santiago — and note that these are the exact platforms where Latino teens spend their day.
NASA and IBM open-source an AI model for the Moon
- The free Lunar Foundation Model, released on Hugging Face, was trained on decades of observations from nine instruments across four missions, and ships with the first open-source lunar dataset: tens of thousands of maps and images (NASA).
- It identifies craters, volcanic formations and likely ice deposits up to 23% better than standard methods, and will support the Artemis program's return to the Moon.
Why it mattersThis is the open-science counterweight to the arms race: any university — including any in Latin America — can download frontier lunar AI today and build on it for free.
The big picture
The same weekend the industry admitted out loud that it might need brakes — Altman's memo, Anthropic's misuse case file, and 25 Fields Medal winners now accusing AI labs of damaging mathematics itself — a state wrote actual rules and public science showed AI's other face. The question is no longer whether AI slows down or gets regulated; it's who sets the speed limit: the labs, the states, or the incidents.
Sources & further reading
- Bloomberg — OpenAI open to slowing cutting-edge AI, Altman tells staff
- Reuters via US News — How Anthropic says Claude was used for weapons, spying and cyber operations
- Bending Spoons — Definitive agreement to acquire Miro for $1.355 billion
- TechCrunch — Bending Spoons to buy Miro for $1.36B, 90% less than its 2022 valuation
- Governor of California — Newsom signs the strongest child safety chatbot and social media laws
- NASA — NASA, IBM launch AI foundation model for lunar science
- TechCrunch — OpenAI's feud with mathematicians is only escalating
- NBC News — Weekend Rundown: AI executives call for a slowdown