Sen. Maria Cantwell of Washington, the ranking Democrat on the Senate Committee on Commerce, Science and Transportation, released a six-point framework under which the National Institute of Standards and Technology (NIST), the federal standards agency, would set safety rules for frontier AI models, per Roll Call. Covered models would face continuous testing by government and third-party groups, with independent audits to confirm compliance. Developers would also have to disclose risks and report safety incidents, including "unsafe recursive self-improvement," in which a model develops itself or other models without human instruction. "To manage risks from advanced AI systems we need clear safety standards, continuous testing, and reporting of serious failures," Cantwell said. The framework carries no legislative text, and a Democratic committee aide said it could become more than one bill.
Read at Roll Call ↗ • Read at The Hill ↗
The Defense Department's Tradewinds program, run by its Chief Digital and Artificial Intelligence Office, accepts product pitches as videos of five minutes or less, Wired reported. Accepted products receive a post-competitive status that lets buyers skip some competition requirements. A panel reviews submissions at least monthly, and a department official said several awards have been made in under a week. OpenAI, Anthropic and Google are listed as participants. This year's focus areas include AI that can "enhance force lethality" and agents that help with joint targeting, including kill chain execution, the sequence from finding a target to striking it. Accepted vendors can more easily receive other transaction agreements, contracts exempt from some federal acquisition rules, and the Pentagon can spend up to $500 million on one without notifying Congress.
Read at Wired ↗
The Finnish Supervisory Agency (LVV) on Tuesday told Tuike Finland, the company representing Google, to suspend site preparation at planned data centers in Muhos and Kajaani by Oct. 23, per Euronews. The agency said trees had been removed, topsoil stripped and site roads built before a required environmental impact assessment was complete. About 330 hectares (815 acres) of forest were felled at Muhos, according to The Guardian. The sites belong to a €13 billion ($15 billion) Google investment to expand its AI capacity, which the company called its biggest in Europe. Tuike must explain how it will proceed by Oct. 14 or face possible enforcement proceedings. Environment Minister Sari Multala said "all companies operating in Finland must comply with the law." Google said it had "fallen short of our own high standards."
Read at The Guardian ↗ • Read at Euronews ↗ • Read at CNBC ↗
Three researchers OpenAI dismissed wrote to the company's board and safety committees asking frontier companies not to pursue work that would weaken monitoring of a model's chain of thought, the step-by-step reasoning text it produces before answering, the Wall Street Journal reported. The letter also asks OpenAI to work with outside safety auditors. "As an industry, we do not yet know how to safely develop and deploy models that we cannot monitor," the letter says, per Gizmodo, which named the authors as Tomek Korbak, Mikita Balesni and Jasmine Wang. OpenAI had said it let the three go for mishandling sensitive company information, as noted in AIPD's October 2nd edition. A company representative said the firings "were not about raising safety concerns or speaking out." An internal memo said OpenAI agreed with the letter's recommendations.
Read at WSJ ↗ • Read at Gizmodo ↗
A consortium of more than half a dozen public and private organizations committed $1.8 billion to produce biology datasets large enough to train AI models that simulate cells, which could stand in for slow and costly lab experiments, SiliconANGLE reported. The Energy Department is the largest contributor at more than $500 million over five years, part of it for microscopy equipment that will gather cell measurements. Biohub, the research nonprofit backed by Mark Zuckerberg, is committing $400 million, three quarters of it for in-house research. The National Institutes of Health will open access to datasets built with more than $500 million in earlier federal funding.
Read at SiliconANGLE ↗
Some 57% of registered voters, a third of Republicans among them, faulted the Trump administration for treating AI risks too lightly or ignoring them, and 54% said the same of Congress, per Reuters. Stricter government regulation of AI drew support from 80% of Democrats and 61% of Republicans. Some 56% backed limits on building data centers, and 57% said the U.S. government should not use AI to choose targets for military strikes, up from 49% in August 2025. A White House official pointed to Trump's new AI task force, which the official said is in charge of ensuring that "America continues to lead the world" on the technology. The online poll of 4,506 adults, including 3,526 registered voters, has a margin of error of 2 percentage points.
Read at Reuters ↗ • Read at Investing.com ↗ • Read at The Hill ↗