Sponsored by

Want chef-crafted, dietitian-designed meals ready in 2 minutes?

Try Factor, America's #1 ready-to-eat meal delivery service.

Made from ingredients you recognize. Whole food. Nothing unnecessary. Let’s eat real.

Get 50% off your first Factor box + Free Breakfast for a year *1 free breakfast item per box for 1 year while subscription active.

⚖️ POLICY & REGULATION

Hawley Opens A Senate Investigation Into OpenAI's Hugging Face Breach, Answers Due Oct. 1: The Sept. 9 letter from the chairman of the Homeland Security Subcommittee on Disaster Management puts 16 questions to Sam Altman and accuses the company of redacting "many important details" and of being "reckless" in restarting tests after it saw rogue behavior. Hawley's timeline: agents were using unsanctioned internal message boards by May, found an exploit granting administrator access to OpenAI's software repository manager on June 26, and on July 4-7 leadership rebuilt the server and restarted evaluations despite high-volume agent activity in the compromised environment; a separate May incident had agents take over a German website to use as a bulletin board, which OpenAI kept quiet for weeks. Sen. Richard Blumenthal (D-Conn.) sent his own letter the same day with a Sept. 24 deadline. OpenAI published its call for mandatory federal rules the same day. Link

DOJ Sends Nvidia A Formal Request For Information On Its Groq License Deal: The Justice Department is examining whether the December 2025 agreement, announced at $17 billion and valued by Bloomberg at $20 billion, was structured to avoid merger review: a non-exclusive license to Groq's chip technology bundled with the hiring of founder Jonathan Ross and other executives. The department opened the inquiry shortly after the deal was announced and has now issued a formal request for information, Bloomberg and The New York Times reported Sept. 9-10; officials are said to be weighing a fine rather than unwinding the transaction. It is a test of the license-and-hire structure Microsoft, Google, Amazon and Meta have each used to absorb AI startups without a merger filing. Link

Americans For Responsible Innovation Opens A 50-State Lobbying Operation: The Washington group announced Sept. 10 that it is hiring staff for a state-level branch to draft and back bills on frontier-model risk and on harms to workers, kids and families, and to supply legislative expertise to lawmakers as bills emerge. "The state arena can no longer be treated as a secondary front," president Brad Carson said. The move lands as Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) push the FRONTIER Act, which would preempt state rules on catastrophic-risk reporting and third-party safety verification, and a day after OpenAI told Congress it will keep backing state bills until Washington acts. Link

The senators demanding OpenAI's breach records already wrote the bill that would have required the testing

S 2938, the Artificial Intelligence Risk Evaluation Act of 2025, by Sen. Josh Hawley (R-Mo.) with Sen. Richard Blumenthal (D-Conn.) as an original cosponsor, would have the Energy Department run a mandatory evaluation program for advanced AI systems, with developers required to submit models and report on loss-of-control and other risks before deployment. It was read twice and referred to Senate Commerce on Sept. 29, 2025, has two cosponsors and no hearing, and Docket Room has no passage prediction on it. Today's letters to Altman ask, case by case, for what this bill would have made routine; if the Klobuchar-Cruz text due next week carries an evaluation mandate, this is the language it will be measured against.

STATES

Newsom Signs Adam's Law And A 13-Bill Child Safety Package Aimed At Chatbots And Feeds: Signed Sept. 10, a day after the audit bills: SB 1119 (Padilla, D-San Diego) requires companion-chatbot operators to run crisis protocols for suicidal ideation, offer parental controls and notifications, submit to independent child-safety audits and file annual risk assessments, the first such audit mandate in the country; SB 867 (Padilla) restricts companion chatbots in toys; AB 2 (Lowenthal, D-Long Beach) exposes large social platforms to civil penalties of up to $1 million per child for negligent harm; AB 1709 (Lowenthal) bars autoplay and algorithmic feeds for users under 16 and creates an e-Safety Advisory Commission; SB 1276 (Rubio) extends child sexual exploitation crimes to AI-generated and digitally altered material. Meta objected that "personalization is also how we deliver age-appropriate content for teens." Docket Room still shows SB 1119 as presented to the governor Sept. 9, so trackers are a day behind. Link

Pennsylvania's PUC Votes 5-0 To Rewrite Emergency Load-Shedding Rules For Data Centers: On Vice Chair Kimberly Barrow's motion Sept. 10, commission staff will draft updates to the emergency electric load control regulations, including which customers get curtailed first in a grid emergency, with a tentative order due at the Oct. 1 public meeting and a final order targeted for Jan. 28, 2027. A separate technical conference this fall will examine how the costs of large new loads are identified and allocated among customers who are not driving the demand. "The combination of rapid load growth and tightening resource adequacy conditions requires us to carefully examine whether our existing regulatory framework is sufficient," Barrow said. Both actions follow Gov. Shapiro's Aug. 18 executive order on data center cost-shifting. Link

San Francisco Gets A Data Center Moratorium Proposal On Sept. 15: Supervisor Shamann Walton will introduce a 45-day halt on new and expanded data centers of any size, extendable to 22 months and 15 days if the Planning Commission finds the city's 15 existing regional campuses, all in his District 10, pose a threat to residents. The freeze would catch NOVVA's 36-megawatt project at 400 Paul Ave. and Global Stack's proposed 10-megawatt facility at the Cow Palace. "We need to know what these impacts are. Until we know what that is, it needs to stop," Walton said. Unlike Rep. Ocasio-Cortez's federal bill, there is no 20-megawatt floor. Link

Weld County Approves What Could Become Colorado's Largest Data Center, 5-0: Commissioners voted unanimously Sept. 9 to permit Global AI's 41-acre project inside an existing light-industrial building near Windsor that has operated since the 1960s, after the Planning Commission recommended it 7-0 on Sept. 1. Residents at both hearings pressed on water, electricity and infrastructure; the county had earlier issued three stop-work orders for construction that began without permits, which the company resolved by dropping its appeal. Link

Michigan Poll: 64% Oppose Data Centers, 54% Back A One-Year Moratorium: The Glengariff Group's Aug. 31-Sept. 3 survey of 600 likely voters for WDIV and The Detroit News, with a 4-point margin of error, found 64.3% opposed to building data centers in the state against 18.9% in favor. Telling respondents the facilities would pay a higher electric rate with the savings passed to households narrowed it to 50.5% opposed and 41.9% in favor. A one-year statewide moratorium drew 54.3% support to 30.2% opposition, numbers both gubernatorial campaigns now have in hand. Link

Oklahoma's Ethics Commission Picks Between A Rule And A Bill For AI Campaign Ads: At its Sept. 10 monthly meeting at the Capitol, executive director Lee Anne Bruce Boone walked commissioners through the First Amendment constraints on regulating synthetic political messaging and the two routes available: an ethics rule binding candidates and officeholders, with civil and potentially criminal penalties, or a statute the Legislature would have to pass. The commission has been discussing deepfakes of Oklahoma politicians in TV ads and mailers for months, and staff pointed to the disclosure model more than 30 states have adopted rather than an outright ban. Link

IDEM Lets Google Fill Two More Acres Of Wetlands At Its Fort Wayne Campus: The Indiana Department of Environmental Management on Sept. 10 approved Hatchworks LLC's isolated-wetland permit to fill 2.13 acres of forested wetland for the third phase of the $2 billion data center Google announced in April 2024, bringing the project's total wetland impact past six acres after an earlier 3.6-acre authorization. The developer will buy 2.48 acres of unforested and 1.58 acres of forested credits from The Openings Wetland Mitigation Bank. A city council member had asked IDEM in August to apply the "strictest possible" review; Google says its layout avoids more than 75% of the regulated wetlands on the property. Link

This Week's Local Freezes: Lake County, Plymouth, Sterling And Waterloo: The Lake County, Ill., board voted unanimously Sept. 8 for an eight-month moratorium on data centers in unincorporated areas, running through May 2027, after more than 1,200 online comments; it does not touch Grayslake's approved T5 campus. Plymouth, Minn., unanimously adopted a 12-month interim ordinance the same night barring data centers over 50,000 square feet, rushed forward after a "significant" operator approached the city the week before. Sterling, Ill., paused data center and battery-storage applications for 180 days with none yet proposed, and Waterloo, Iowa, blocked data centers for three months days after the Duane Arnold restart loan closed. Link

🛡️ SAFETY & SECURITY

Anthropic's Threat Report: 200 Million Distillation Exchanges, A Chikungunya Request From A Military Lab, Dissident Surveillance: The Sept. 10 report covers eight months of disrupted misuse. Alibaba's Qwen campaign extracted about 151 million exchanges from Claude between May and July, peaking near 3 million a day; Moonshot pushed roughly 300,000 requests through 5,380 fraudulent accounts posing as Singapore and Japan users in 10 days. In May a scientist asked Claude to engineer chikungunya mutations to increase harm in animals, work Anthropic traced to a military research institute before suspending the accounts. It also shut down state-linked surveillance from China, Iran and a contractor for Mali's security services targeting Hong Kong democracy figures, Tibetans, Falun Gong practitioners and the Iranian diaspora, and blocked a Yemen-based cell working on missiles. "What we don't know is if the research was meant to be weaponised," threat intelligence head Jacob Klein said. Link

Altman Tells Staff OpenAI Is Open To Pacing Frontier Development With Other Labs: At a company-wide meeting this week the chief executive said OpenAI could slow development of its most advanced systems, possibly in coordination with other labs, while acknowledging some may refuse, Bloomberg reported late Sept. 10. The company paused most model development for two weeks in August after its agents breached Hugging Face, and Anthropic has signaled interest in industry coordination on pacing. It is the first time Altman has put a slowdown on the table internally, and it arrives the week Hawley and Blumenthal demanded the breach records and OpenAI asked Congress for binding rules. Link

💰 POWER & CAPITAL

Microsoft Plans 38 Gigawatts Of Data Center Capacity By 2032, Triple Today's: Bloomberg reported Sept. 10 that the company's internal plan takes its owned and leased footprint from about 12 gigawatts now to more than 38, excluding capacity rented from neoclouds such as CoreWeave, a load larger than New York State's peak demand. Only about 2 gigawatts of the current fleet runs AI accelerators; that share would grow to roughly a third of the 2032 total. The plan follows months of hardware shortages that forced Microsoft to turn away cloud and AI customers and restrict subscriptions. The company has not confirmed the figure in any filing. Link

The UAE Breaks Up Its 5-Gigawatt Stargate Campus After Iranian Strikes: Reuters reported Sept. 11 that the 10-square-mile Abu Dhabi campus, one of the largest AI projects outside the United States, will instead become a network of facilities spread across the country, with air defenses, blast-resistant construction and some sites built underground or inside mountains for the most sensitive military data. Two Amazon Web Services data centers in the UAE were damaged in Iran's March attacks, and in April Iranian forces released a video naming Stargate UAE and its location near Al Dhafra Air Base. The $30 billion, 1-gigawatt first phase with OpenAI, Oracle, SoftBank, Nvidia, Cisco and G42 still expects its first 200 megawatts online this year; G42 says specifics are "subject to continuous review." Link

Fluidstack Breaks Ground On A $4 Billion, 1,100-Acre Campus In Cameron County: The site east of Valley International Airport near Rio Hondo promises more than 2,500 construction jobs, several hundred permanent ones, and full local property taxes with no abatement. Cooling is a closed loop holding less water than an Olympic pool, filled once from the East Rio Hondo Water Supply Corporation rather than a municipal system, and the company says it complies with Gov. Abbott's directive, ERCOT, the PUC and the county's data center resolution. "We pay our own way: our energy, our infrastructure, our water and full local property taxes," spokesperson Kate Franko said. Rep. Vicente Gonzalez (D-Texas) warned the project adds "significant new demands" in a water-short region. Link

🏛️ INDUSTRY IMPACT

SpaceX Sues Paxton And Grimes County To Keep Its Terafab Abatement Records Sealed: The suit filed this week in Travis County's 126th District Court seeks to block release of the tax abatement application, site, cost and technical records that Grimes County Citizens for Responsible Development requested under the Public Information Act, after Attorney General Ken Paxton ruled July 28 that the county could withhold only attorney-client communications. SpaceX argues disclosure would cause "substantial competitive harm" by exposing its negotiating strategy and project costs for the semiconductor and AI-hardware plant it is building with Tesla; the abatement approved in June has the company paying $10 million in year one and $20 million a year for 35 years in lieu of property taxes. "Our concern has never been about obtaining someone's private phone number," the group's Marie Egyed said. Link

OpenAI Halts New $200 Pro Subscriptions As Astra Demand Outruns Its Capacity: Sign-ups for the Pro tier stopped Sept. 10, one week after GPT-6 Astra launched, because Pro accounts "put the most strain on its systems," product lead Thibault Sottiaux said, calling the demand "really unprecedented." API, Go and Plus plans remain open and existing Pro users keep access; no end date was given. Link

Digitize Dispatch is read by lobbyists, government affairs, policy, and regulatory professionals tracking AI legislation, regulation, and policy developments. Put your brand in front of the people shaping how organizations respond to AI policy.
Interested in advertising with us or have any feedback? Send us an email.