26 attorneys general demand federal AI safety rules from Congress

Share
26 attorneys general demand federal AI safety rules from Congress

State attorneys general are done waiting for Congress to act on its own — and Microsoft has quietly retired the label it invented to sell AI laptops. Three stories for you this afternoon.


26 attorneys general from 24 states have written to Congressional leaders demanding federal AI safety rules. The letter, sent Wednesday to Speaker Mike Johnson, Senate Majority Leader John Thune, and Democratic leaders Hakeem Jeffries and Chuck Schumer, asks for four concrete things: independent safety testing of frontier models instead of uniform federal benchmarks, government-led incident investigations with public reporting, internal safety staff empowered to halt deployments, and international coordination on pacing — all while preserving states' own enforcement power, meaning no preemption of existing state law. It also carries a sharp line: when autonomous agents are directed at harmful activity, "they violate criminal law." New York's Letitia James led the coalition, with Oregon's Dan Rayfield among the signatories; the states span both parties, from California and Massachusetts to Oklahoma and North Carolina.

The timing is the story. Congress is already weighing a duty-of-care law that could make labs responsible for unsafe models — as we covered in Senate weighs a duty-of-care law that could block unsafe AI models — and yesterday's letter is the states putting a marker down: they want federal floors, not federal ceilings that erase what California, Colorado, and New York have already built. That no-preemption demand is the real fight. Washington has a long habit of settling AI federalism by preempting the states, and 26 attorneys general saying "set standards, don't disarm us" is the clearest signal yet that the states will fight it. What to watch: whether Johnson — who has said AI safety is the companies' job, not Congress's — acknowledges the letter at all.


Microsoft has stopped calling its new Surface laptops "Copilot+ PCs" — even though they qualify. The Surface Pro 12-inch and Surface Laptop 13-inch, both shipping October 13 with Qualcomm Snapdragon X2 Plus chips and an 80 TOPS neural processor, meet every requirement of the Copilot+ bar (16GB of RAM, 256GB of storage, a 40+ TOPS NPU). Surface corporate vice president Brett Ostrum confirmed the branding decision outright: "They do meet all the requirements of our previous bar for what Copilot+ devices are." The Copilot+ landing page now redirects to a generic "performance PCs" page.

The label did its job and then wore out its welcome. It sold the idea of the AI PC during a post-COVID sales lull — Copilot+ machines took 15% of US premium laptop sales in the 2024 holiday season, by Satya Nadella's count — but the Recall privacy backlash tarnished it, Dell already killed its AI PC branding, and IDC's Jitesh Ubrani says interest has been "wavering for a while" now that cloud models are everywhere and on-device use cases stay thin. Expect Nvidia's incoming RTX Spark — including a Surface Laptop Ultra — to get the marketing budget instead.


Another DeepMind engineer is leaving — and this one built the chips that make AI cheaper. Robert O'Callahan, a technical expert at Google DeepMind in Auckland who worked on chip design tools that cut the cost and latency of AI, wrote Wednesday that "aiming for ASI in the near future is inherently irresponsible" and that AI's "current rate of change is far too high." He rates the risk from superintelligent AI "real but uncertain" but says he'd bet the harms outweigh the benefits, and he wants to do "unambiguously pro-human" work instead. He is explicit that his exit is not part of the viral wave of safety resignations — we covered DeepMind's first high-profile departure earlier this month in Google DeepMind safety researcher quits, says AI 'could kill us all'. O'Callahan's departure matters for a different reason: it's a builder, not a safety researcher, concluding that the infrastructure work itself can no longer be justified.

Should Congress write federal AI safety rules the states can't weaken — or is a messy patchwork better than nothing? Tell us in the comments.

Read more

Mistral unveils Le Chonk: a 1T-parameter open-weights model

Mistral unveils Le Chonk: a 1T-parameter open-weights model

The biggest open-weight release outside China lands in public preview today, and the country that spent the week promising its own frontier model just put a price on the ambition. Mistral has opened a public preview of Mistral Large 4 — codenamed "le Chonk" — a 1 trillion-parameter mixture-of-experts model with 49 billion active parameters, natively multimodal, which the company calls its largest and most capable model to date. The preview API is live today on Mistral Studio at $1.36 per milli

Google signs 3.6 GW power deal, a quarter of it new nuclear

Google signs 3.6 GW power deal, a quarter of it new nuclear

Grid capacity, not chips, is becoming the binding constraint on the AI buildout — and on the same day, the labs told an Australian inquiry they can live with mandatory incident reporting. Google has contracted 3.6 GW of power from Constellation Energy across the PJM grid — the largest electricity deal in the region's history, and the biggest single power commitment any AI company has made. The agreement covers 3,590 megawatts over 13 states, with 890 megawatts of new nuclear capacity coming fr

Korea probes AI agents in bank hacks as president cites 'signs'

Korea probes AI agents in bank hacks as president cites 'signs'

South Korea opened a formal investigation into whether AI agents drove a wave of bank breaches — and it isn't the only AI story moving money today. President Lee Jae Myung said "signs" point at AI models, and the probe is now at the highest level a national banking sector has seen. Speaking at a cabinet meeting, Lee said that "in some hacking incidents, signs have emerged of AI being used, causing considerable public concern and anxiety," and police have since opened a full-scale investigation

Open Source Radar — October 6: nothing leaves your machine

Open Source Radar — October 6: nothing leaves your machine

Today's trending board is all projects we ran earlier this week, so the fresh signal comes from the Product Hunt launch slate instead — three open-source tools that share one instinct: your phone, your pixels and your MCP traffic should stay on hardware you control. All three verified at the source. iphone-use (Rust, MIT, about 59 stars) is computer-use for a real iPhone: an agent reads the screen as text, taps, swipes and types over WebDriverAgent, and every action comes back with an honest v