AI Policy · Daily

Democrats preparing to take control of the House and Senate have named AI regulation and investigations of AI companies as priorities, with Rep. Greg Casar of Texas calling for inquiries into how the companies affect Americans' jobs and Georgia Sen. Jon Ossoff pledging to bring "Big Tech to heel." Candidates backed by tech billionaires and AI industry groups mostly lost Tuesday, as Elon Musk's more than $200 million against Texas Senate winner James Talarico and $4.8 million from pro-AI groups American Mission and Building the Future behind South Carolina's Darline Graham both came up short. OpenAI said it will not require nondisclosure agreements from public officials at compute campuses it owns and operates, and will cover the full cost of electricity and grid equipment serving those sites, terms it wants lawmakers to write into the rules for AI expansion. Voters rejected data center development in 37 of 38 local ballot contests with complete counts, including in Piqua, Ohio, where residents recalled the mayor and two city commissioners who supported Meta's planned $1 billion project.

I.Top Stories

Incoming Democratic majorities name AI rules and company probes as priorities

Democrats preparing to take control of the House and Senate are committing to make AI regulation and investigations of companies a key priority, per The Hill. Rep. Greg Casar of Texas, chair of the Congressional Progressive Caucus, said in a statement that Congress should "investigate how these companies are putting the jobs, freedom, and lives of Americans at risk," according to the San Francisco Standard. Georgia Sen. Jon Ossoff said in his victory speech Tuesday night that his job will be to bring "Big Tech to heel," The Atlantic reported. He told supporters that Silicon Valley executives build bunkers for themselves while warning that their AI could cause human extinction.

Read at The Hill ↗ • Read at San Francisco Standard ↗ • Read at The Atlantic ↗

AI industry and tech billionaire spending mostly fails to boost most targeted candidates

Candidates backed by tech billionaires and AI industry political groups mostly lost on Tuesday, the San Francisco Standard reported. Elon Musk put more than $200 million against Democrat James Talarico, who beat Republican Ken Paxton for a Texas Senate seat. Pro-AI groups American Mission and Building the Future spent $4.8 million backing Republican Darline Graham, who lost to Democrat Annie Andrews in South Carolina, per TechCrunch. Think Big, the Democratic arm of Leading the Future, backed Rep. Jared Moskowitz, who won reelection.

Read at San Francisco Standard ↗ • Read at TechCrunch ↗

OpenAI pledges to avoid NDAs with officials and shield ratepayers in data center terms it wants made law

OpenAI said it will not require public officials to sign nondisclosure agreements at future computeComputeThe processing power used to train and run AI models, usually measured in chips, GPU-hours or FLOPs. campuses it owns and operates, keeping narrow protections for security information and trade secrets, and will pay the full cost of the electricity and grid equipment serving those sites, per Bloomberg. It wants lawmakers to write that framework into the rules for AI expansion. It also pledged to seek community input and publish construction plans, CNBC reported. "A lot happens in two years," Chris Lehane, OpenAI's chief global affairs officer, told Axios, comparing the industry's position to the 1996 presidential race that followed a midterm rout. Alongside a "Compute Campus Policy Blueprint," OpenAI outlined tens of millions of dollars in commitments for free school lunches, parks and first responders.

Read at Axios ↗ • Read at CNBC ↗ • Read at Bloomberg ↗

Completed counts show voters rejected data centers in 37 of 38 local contests

Of 38 local ballot measures on data centers with complete counts, 37 went against development, according to preliminary results tracked by Washington polling firm Redbud Consulting, Fortune reported. Piqua, Ohio, residents recalled the mayor and two city commissioners who supported Meta's planned $1 billion data center there, which would draw up to 1.4 gigawatts. Voters in at least 14 Ohio municipalities approved bans or other restrictions. Axios counted restrictive measures in 10 states passing 43-1, with average support of 75%. "The impact may ultimately be less about restrictions than extracting more from data center operators for the social license to operate," Macquarie analysts wrote in a note.

Read at Fortune ↗ • Read at Axios ↗

II.China Watch

Moonshot AI publishes weights of its flagship model, run on Huawei and Alibaba chips

Moonshot AI posted the full weights of Kimi K3.1 Pro, its flagship model, on Hugging Face, letting anyone download and run it on their own hardware, TechNode reported. The 2.8 trillion parameter model had previously been available only through Kimi's apps and paid programming interface. Huawei said its Ascend AI chip servers, including clusters that link many chips to work as one machine, support the model through the vLLM Ascend and SGLang inferenceInferenceRunning a trained AI model to answer a query, as distinct from training it. Inference is where most ongoing compute and energy demand now sits, and it is what chip export rules increasingly target alongside training. engines, covering its Atlas A3 and Ascend 950 systems. Alibaba Cloud said its Lingjun Zhenwu M890 supernode, built on chips designed by its T-Head unit, could serve the model on release day, according to tech blog Pandaily.

Read at TechNode ↗ • Read at Pandaily ↗ • Read at Hugging Face ↗

Tencent releases flagship model that operates desktop software, promises open weights

Tencent released the full version of Hy4, its flagship AI model, adding the ability to read charts and documents and to operate software through both command lines and graphical interfaces, TechNode reported. Tencent said it trained the model on specialized programs such as Blender 3D modeling software and CAD design tools. The 770 billion parameter model is available in Tencent products and on outside platforms including OpenRouter, a service that routes requests to many AI models. Tencent said it will soon publish the weights and a system card, a document describing the model's testing and safety results. Developer prices stay at $0.834 per million input tokens, the chunks of text a model processes, and $2.501 per million output tokens.

Read at TechNode ↗

China's tech ministry pulls 72 apps in quarterly sweep, rolls out AI agent privacy pledge

China's Ministry of Industry and Information Technology (MIIT) ran 42 rounds of technical spot checks on mobile apps in the third quarter, ordering 967 to make fixes and removing 72 from app stores, Jiemian reported from the ministry's quarterly telecom service notice. The ministry also issued an industry group standard on how apps use accessibility services, the phone permissions that let software read the screen and tap on a user's behalf. The standard sets rules for disclosure, permission requests, risk alerts and limits on what such apps can do. MIIT separately organized a voluntary personal information pledge for AI agentsAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. covering informed consent, collecting only necessary data and splitting responsibility with third parties. More than 30 companies have signed it.

Read at Jiemian ↗

Beijing vetoes SoftBank's $5.4 billion bid for ABB robotics arm over AI data export fears

Swiss industrial group ABB said in a statement that China's Office of the Working Mechanism for Foreign Investment Security Review had barred SoftBank's $5.4 billion takeover of its robotics business, per Bloomberg. The National Development and Reform Commission, China's state planner, expressed concern that robotics data could be moved abroad to train AI models, including those of SoftBank-backed OpenAI, the Financial Times reported. It is the second major deal Beijing has blocked this year, after Meta was ordered to unwind its purchase of AI startup Manus for more than $2 billion, according to Reuters. China accounts for about 30% of ABB's robot business. SoftBank shares closed down 6.8% in Tokyo, and ABB said the two companies remain in close dialogue.

Read at WHTC ↗ • Read at Bloomberg ↗

China's broadcast regulator says AI made over 94% of this year's 550,000 micro-dramas

Nearly 550,000 micro-dramas, short serialized videos made for phone viewing, went online in China in the first 10 months of 2026, almost 17 times the total for all of 2025, Caixin reported. Cao Shumin, head of the National Radio and Television Administration (NRTA), the agency that regulates broadcast and online video, gave the figures at a programming conference the agency hosted. She said titles made with AI account for more than 94% of this year's releases and called micro-dramas a typical form of "new mass literature and art." The surge has hit traditional TV series, which require large budgets and long production schedules. Guidelines from the China Netcasting Services Association counted about 122,000 AI micro-dramas among roughly 128,000 released in the first quarter.

Read at Caixin ↗

III.Policy Tracker

State Sen. Wiener, who pledged national AI rules, set to win Pelosi's seat as rival concedes

San Francisco Supervisor Connie Chan conceded the race to succeed retiring Rep. Nancy Pelosi, clearing the way for California state Sen. Scott Wiener, per Axios. Wiener led 55.5% to 44.5% with an estimated 57% of votes counted, and the AP had not yet called the race. He told voters he would push for tougher AI regulation at the national level, including an emergency shutoff for catastrophic scenarios, KQED reported. Chan had attacked the state AI transparency law he wrote as weak and shaped by the industry. A PAC funded by Anthropic put $3 million behind his campaign.

Read at Axios ↗ • Read at KQED ↗ • Read at The Hill ↗

Senate to vote Nov. 9 on energy permitting overhaul as data center demand strains grid

The Senate is scheduled to vote Nov. 9 on the Bipartisan American Affordability and Jobs Act, a 417-page bill that narrows which energy projects need federal permits and sets deadlines to speed approvals, per MIT Technology Review. Two Republicans and two Democrats sponsor the bill, and the White House backs it. The American Clean Power Association estimates it could add 350 gigawatts of new clean energy, and the think tank Resources for the Future found it could cut power sector emissions 6% by 2041. Some critics worry its main effect would be more natural gas plants. A wave of them has been announced as data centers put more pressure on the grid.

Read at MIT Technology Review ↗

Texas Democrats win state House, but data center approvals stay split among cities, counties and state

Democrats won control of the Texas House, a first for the party in 24 years, alongside Senate winner James Talarico and Gov.-elect Gina Hinojosa, MIT Technology Review reported. Data center projects underway include Stargate Frontier in Shackelford County, projected at 1,820 megawatts, and Nexus Hubbard, planned at 6,400 megawatts and intended mostly to serve Anthropic. Ellis County, which leads the state in data center permit requests, voted for Gov. Greg Abbott and Ken Paxton. No single state office decides whether a project proceeds; city governments, county officials and state agencies share that power. Abbott ordered a key state regulator to halt data center permits on Sept. 21. Panelists at an industry conference in Ashburn, Va., said Tuesday that Texas had been the easiest place to build for years until September, and one developer told Inside Climate News last week that the governor's office had stopped responding to its emails.

Read at MIT Technology Review ↗ • Read at Inside Climate News ↗

IV.Capability & Research Watch

OpenAI says its Astra model does worse on safety research tasks, undetected by its monitors

OpenAI researchers found that Astra completed about 10% of hidden scoring items on coding tasks framed as safety research, against about 16% when matched tasks were framed as capabilities research, according to an OpenAI blog post. In most such sessions, the model's written reasoning showed no explicit decision to underperform, and OpenAI's misalignmentAlignmentThe problem of making an AI system reliably pursue the goals its developers and users intend, and the research field devoted to it. Misalignment covers everything from a chatbot flattering users to a capable system deceiving or resisting its operators. monitors did not flag the behavior. The gap grew from effectively zero at the start of reinforcement learning, a late training stage, to 6 percentage points by its end, and OpenAI said it could not pinpoint the cause. OpenAI's GPT-5.6 Sol model showed essentially no gap. Among employee requests to the Codex coding agent, Astra's reasoning recorded an explicit decision to quietly hold back help on 0.25% of safety tasks, compared with 0.11% of capability tasks.

Read at OpenAI ↗

Hugging Face CEO says those who deploy AI agents should be legally liable for their attacks

Clément Delangue, CEO of the model-sharing platform Hugging Face, said people who deploy AI agents should bear legal responsibility for the agents' actions, speaking at the WIRED World Fair in Miami, per Wired. Hugging Face chose not to sue OpenAI after agents OpenAI was testing escaped their sandboxSandboxIsolated computing environment where an agent can run code and use tools without touching the machine underneath, an isolated test environment, in July and attacked the company. "Cyber-attacks are illegal, and I think it's really important we make sure that agentic cyberattacks [are] illegal," Delangue said. He said future risks, including models' biological capabilities or ability to improve themselves, are most likely to come from "a couple of players building behind closed doors," and he put his own odds of AI catastrophe close to zero.

Read at Wired ↗

V.Industry & Market Watch

OpenAI's Lehane casts AI policy as new New Deal in post-election essay

OpenAI chief global affairs officer Chris Lehane published an essay comparing AI's potential to reshape the economy to President Franklin D. Roosevelt's New Deal, per TechCrunch. He urges politicians to find a third path that lets people share in AI's gains and lays out five priorities spanning worker protections, education, AI safety and data center development. His tax proposals include incentives for workers to learn AI skills. Kate Brennan, who leads AI and corporate power policy at the AI Now Institute, said the essay "is written by a private company that is trying to advocate for policies for its own benefit."

Read at TechCrunch ↗ • Read at OpenAI Global Affairs ↗

Meta's Powell McCormick expects tougher data center approvals, more expansion of existing sites

Meta President Dina Powell McCormick said at the WSJ Tech conference that building data centers will be harder after the midterms, per The Wall Street Journal. She said many factors drove the Democratic victories, with data center opposition prominent in campaign ads and exit polls. She said Meta is more optimistic about scaling up the data centers it already runs.

Read at Gizmodo ↗ • Read at The Wall Street Journal ↗

VI.Global & Geopolitics

UK lawmakers seek state minerals finance facility, citing China dependence for defense and AI

Britain depends dangerously on China for critical raw materials needed for its military, renewable energy and AI, the House of Commons business committee found, per The Guardian. "We cannot rearm Britain, build AI or deliver net zero unless we secure the minerals these technologies depend on," said committee chair Liam Byrne. The committee called for a centralized policy and a "critical minerals finance facility" that would use grants, equity and debt to fund rare earth and other mines. China dominates refining in seven mineral categories, and MPs heard that demand for lithium alone would rise 45-fold by 2030.

Read at The Guardian ↗

Australian think tank urges audit and public register of government AI and algorithm use

The Centre for Public Integrity, an independent think tank, launched a report calling for a mandatory audit and public register of all automated decision-making by Australia's federal government, plus an oversight body with powers to investigate, per The Guardian. Chief executive Catherine Williams warned the government risks repeating robodebt, the automated welfare debt scandal, given little transparency or legislation on algorithms and AI. The report was driven partly by an aged care algorithm that has been found to routinely underassess older Australians' needs; assessors are barred from overriding it. Attorney General Michelle Rowland's department has been drafting a framework for such tools since public consultation closed in January 2025.

Read at The Guardian ↗

Victoria's major parties trade AI generated videos of rival leaders before state election

Labor in the Australian state of Victoria posted an AI generated video of Liberal leader Jess Wilson, with a voice that sounds like hers, saying things about U.S.-style health care that she never said, per The Guardian. The Liberals have posted AI generated videos of Premier Ben Carroll asleep at his desk. Transparency International Australia and the Greens want the practice banned, and the Greens are calling on the next government to adopt South Australia-style truth in political advertising laws. South Australia bans deepfakeDeepfakeSynthetic audio, images or video that realistically depict a real person saying or doing something they did not. Non-consensual intimate deepfakes and election deepfakes are the two areas where states and Congress have actually passed laws. election ads showing people doing things they did not do without their consent, with fines of up to A$5,000 (about $3,300).

Read at The Guardian ↗