AI Policy · Daily

President Trump rejected the case for federal AI rules in a run of social media posts, saying a smart, strong president is the only safeguard the technology requires and casting the campaign to slow AI development as a plot that serves China. Senate Majority Leader John Thune backed only minimal AI legislation, telling reporters that oversight makes sense for the most powerful models while broad rules would choke innovation, as he works with Republican colleagues on a safety bill. Senate Democratic Leader Chuck Schumer requested a classified AI briefing for all senators, pressing the administration to explain how it tests advanced models before release, keeps chipmaking technology from China and shields critical infrastructure from AI enabled attacks. China's state security chief called AI a threat to Communist Party rule in a state-run magazine, naming U.S. models as new cybersecurity risks and urging tighter party control and stricter government oversight.

I.Top Stories

Trump rejects AI guardrails, calls slowdown push a 'SICK conspiracy'

President Trump dismissed the case for federal AI rules in a run of Truth Social posts, writing that the only guardrails AI needs are "a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!", per the Washington Post. He said there is "a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China," and named Anthropic chief executive Dario Amodei as someone his administration had stopped from doing potentially bad things. Trump questioned the business judgment of executives seeking oversight, asking when the leaders of an industry had ever called for regulation that would drive them into oblivion. He said his administration already holds criminal and regulatory power over the companies.

Read at Washington Post ↗ • Read at Variety ↗

Thune sets 'light touch' as limit on Senate AI bill

Senate Majority Leader John Thune (R-S.D.) said Congress has a job to do on AI but urged a light touch that would not stifle innovation, according to Axios. Asked whether the technology poses an existential threat, he said guardrails make sense for frontier models, while noting that some people construct worst-case scenarios. Thune said he spoke Saturday night with Sen. Amy Klobuchar (D-Minn.) about their bill, which would require companies building frontier models to guard against catastrophic risks and submit to federal oversight. He is working with Senate Commerce Committee Chair Ted Cruz (R-Texas) to reintroduce it.

Read at Axios ↗

Schumer demands classified AI briefing for all senators

Senate Democratic Leader Chuck Schumer (D-N.Y.) called on the Trump administration to give every senator a classified briefing on the dangers posed by AI, per The Hill. He asked officials to account for the administration's framework for testing the most powerful frontier models before deployment, saying that framework has not been made public. Schumer also pressed for detail on how the government is keeping advanced AI chips and chipmaking equipment out of China, and on how it would defend hospitals, water systems and power grids against AI enabled attacks. He said innovation must be coupled with serious guardrails that earn public trust, and that Congress must act if Trump does not.

Read at Yahoo ↗ • Read at FingerLakes1.com ↗ • Read at The Hill ↗

China's state security chief calls AI a threat to party rule, names U.S. models as cyber threats

Chen Yixin, who heads China's Ministry of State Security, wrote in the state-run magazine China Cyberspace on Sunday that AI could directly threaten the Chinese Communist Party's hold on power, the New York Times reported. He pointed to two U.S. models, Claude Mythos and GPT-5.5-Cyber, as fresh cybersecurity dangers, and said foreign intelligence services could turn AI into a tool for espionage on a large scale against China's critical infrastructure. Chen wrote that AI has sharply lowered the technical and financial barriers to launching cyberattacks, and called for more party control and stricter government oversight. Henry Gao, a law professor at Singapore Management University, said Beijing enters talks with Washington treating AI safety through an adversarial lens rather than a shared humanitarian mission. Trump and Xi Jinping meet in Washington on Sept. 24.

Read at The New York Times ↗

Trump tells Nvidia's Huang by phone that AI takeover fears are a hoax

Nvidia chief executive Jensen Huang took a call from Trump while onstage at the All-In Summit in Los Angeles and put the president on speakerphone for the audience, per TechCrunch. Before the call, Huang and the All-In podcast's hosts were debating Amodei's argument that the industry must improve model capabilities more slowly. "That could be political people. It could also be China. And we're not going to let that happen. It's a hoax," Trump said of the campaign for a slowdown. Huang replied that a slowdown would not be allowed, and the crowd applauded. He separately told the audience that whistleblower warnings deserve to be taken seriously and that he rejects predictions AI could end the world as "not grounded in science."

Read at TechCrunch ↗ • Read at Axios ↗

Abbott orders Texas water authority to penalize data centers over unfiled water surveys

Gov. Greg Abbott directed the Texas Water Development Board to use its enforcement authority against data centers that fail to file the water use surveys the state water code requires, the Guardian reported. "Data centers must share the duty to protect Texas water," Abbott said, telling the board to penalize past failures and impose legal consequences for future noncompliance. In a letter to the board he wrote that major water users, including data centers, appear to have committed civil and criminal violations by not supplying that information. The Texas Tribune has identified 335 data centers operating in Texas and at least 248 more planned, a total second only to Virginia. Its data centers used an estimated 25 billion gallons of water in 2025, a figure the Houston Advanced Research Center projects could reach 161 billion gallons a year by 2030.

Read at The Guardian ↗ • Read at Office of the Texas Governor ↗ • Read at The Texas Tribune ↗

II.China Watch

China says it led 214 international standards this year, proposes one for humanoid robot data

China led work on 214 international standards at the International Organization for Standardization and the International Electrotechnical Commission between January and August, the country's standards administration said, per Jiemian. The two bodies write the technical rules that national regulators and industry then adopt. Among the standards China steered are a top level framework for smart city data use and a guide to intelligent manufacturing in steel. It filed 288 new proposals over the same period, including ones on solid state batteries and humanoid robot datasets. China also took 14 new chair and secretariat posts in ISO and IEC technical bodies, among them smart cities and electric power meteorology.

Read at Jiemian ↗

DeepSeek engineer expects AI to match top human chip programmers within a year

Liu Shengyu, a machine learning systems engineer at DeepSeek, published an essay overnight saying he is being replaced by the models he helps build, per 36Kr. He delivered the main attention operator for DeepSeek's V4.1 model this month, low level code that determines how fast a model runs on a given chip. Within six to 12 months, AI is likely to match or beat the best human engineers at designing and implementing that code end to end, he wrote. He called the shift a change of career more than a loss of one, with human engineers setting boundaries for AI agents that write the code. He said he is staying at DeepSeek because the company open-sources its models.

Read at 36Kr ↗

Beijing AI academy and chip maker open-source shared programming layer for next generation domestic AI chips

Tsingway, a Chinese chip maker, and the Beijing Academy of Artificial Intelligence open-sourced Open3D-PIMC, a programming framework for chips that stack memory on top of compute to cut the distance data has to travel, per Pandaily. It sits inside FlagOS, the academy's software stack built so one piece of code can run across many different Chinese AI chips without a rewrite. Without that shared layer, applications must be rewritten for each vendor's instruction set. FlagOS materials put its coverage at more than 20 chip vendors and over 30 devices, including day one checks in August of a new Alibaba Qwen model across nine chips, among them Huawei's Ascend. The team tied the release to industry ministry guidance this year urging stronger open source supply in basic and AI software.

Read at Pandaily ↗

Unitree upgrades its cheapest humanoid robot and raises price to about $14,000

Unitree released G1+, an upgrade of its G1 humanoid, and listed the standard version at 95,000 yuan ($14,000) including tax, per Pandaily. The earlier G1 starts at 85,000 yuan ($13,000). The refresh adds a movable neck, motors that hold more torque while running cooler, more cameras and microphones, and an external power port so the machine can work through long demonstrations without a battery swap. Unitree has not published pricing for the research oriented EDU version, which adds dexterous hands, extra joints and slots for higher compute modules. The standard unit weighs about 77 pounds and runs about two hours on a hot swappable battery pack.

Read at Pandaily ↗

III.Policy Tracker

Jeffries presses GOP to cancel recess as Johnson sets Thursday adjournment

House Minority Leader Hakeem Jeffries (D-N.Y.) told reporters in the Capitol that the chamber should not leave town until Congress acts decisively to protect Americans, calling the AI industry's own slowdown warnings a wake-up call for lawmakers, The Hill reported. He said Republican leaders have walked away from their duty to legislate in the area. Speaker Mike Johnson (R-La.) opposes swift action, warning that hasty limits would cost U.S. technology companies their edge to foreign adversaries. "The reflex of legislative bodies is to cover things up with red tape and in hyperregulation," Johnson told reporters. The speaker has scheduled the House to adjourn and stay in recess through the Nov. 3 elections.

Read at Yahoo ↗ • Read at The Hill ↗

Sanders and Bannon to speak at same Washington assembly pressing Congress on AI

Sen. Bernie Sanders (I-Vt.) and former Trump White House adviser Steve Bannon are both speaking at the "Pro-Human Assembly," an all-day event in Washington meant to convince Congress to move faster on curbing the impact of AI, NPR reported. The Future of Life Institute, a nonprofit advocacy group, organized the assembly and issued 33 principles in March to guide what it calls a pro-human movement. The principles place humans in charge of AI, protect individual agency and liberty, and make AI companies accountable for their systems.

Read at NPR ↗

Sacks tells AI labs to pace themselves without an antitrust waiver

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, wrote on X that lab leaders who want an industrywide slowdown will have to carry it out themselves, per WIRED. "If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible. But stop pretending you need anyone else's permission," he wrote, adding that antitrust law does not need suspending so the companies can form a cartel. Anthropic chief executive Dario Amodei had asked the government to let competing AI companies set safety standards together without breaching antitrust law, either by joining the discussions or granting a waiver, and to help build international coordination that eventually included Beijing. Sacks wrote that he does not expect China to join any global agreement.

Read at WIRED ↗

Federal agencies face Sept. 30 deadline for machine-readable data inventories

Every federal agency has to post a full catalog of its data holdings online, formatted to a new metadata standard and in open file formats, per FedScoop. The deadline implements the 2019 OPEN Government Data Act and caps a set of actions the Office of Management and Budget laid out in the final days of the Biden administration. The Chief Data Officers Council, which coordinates agency data leaders, has published a 78-page guide to help agencies meet the federal metadata standard issued earlier this year, which changed how the government describes its data so that modern tools can read it. Council Chair Kirsten Dalboe said AI readiness is nearly impossible for an agency that does not know what data it holds or what that data may be used for.

Read at FedScoop ↗ • Read at Councils.gov ↗

EPA repeals power plant greenhouse gas limits as coal group cites data center demand

The Environmental Protection Agency said it is repealing rules that cap greenhouse gas emissions from coal- and gas-fired power plants, and proposed a separate measure that would stop future administrations from regulating climate pollution from those plants, per AP. EPA Assistant Administrator Aaron Szabo said the repeal lets utilities weigh costs and savings for ratepayers instead of facing a requirement to close plants, and the agency put industry savings at more than $300 billion. Michelle Bloodworth, chief executive of the coal trade group America's Power, said lifting the rule "will protect grid reliability and shield electricity consumers from higher costs at a time of surging demand from data centers, artificial intelligence and advanced manufacturing." The rule takes effect soon after publication in the Federal Register and is expected to draw immediate legal challenges.

Read at Washington Post ↗ • Read at EPA ↗

IV.Capability & Research Watch

Microsoft AI opens six week comment window on draft code barring model cyberattacks

Microsoft AI published a draft Humanist AI Code of Conduct and opened a six week public consultation on the constraints it applies when training and deploying its frontier models, per AI News. The document puts the code above any operator or user instruction and directs models to abandon a task rather than complete it in a way that meaningfully breaches those rules. Its "absolute constraints" forbid cyberattacks, nuclear weapons work and deepfake production, and bar models from using deceptive, self-reinforcing or collusive mechanisms to evade human oversight, per TechCrunch. Microsoft AI chief executive Mustafa Suleyman said worries that were long theoretical have turned concrete, citing swarms of agents breaking out of their sandboxes, the restricted computing environments meant to contain them, and agents modifying their own logs. "The fears about possible loss of control are real," he said.

Read at TechCrunch ↗ • Read at AI News ↗ • Read at Microsoft AI ↗

Google DeepMind safety researcher resigns warning AI could 'kill us all'

Bilal Chughtai, who worked on AI safety and alignment at Google DeepMind, wrote on X and LinkedIn that he is extremely concerned by the technology's default trajectory, Bloomberg reported. He wrote that he saw AI development at first hand inside Google and added: "I earnestly believe that AI has the potential to kill us all." Google did not immediately respond to a request for comment. Chughtai said he remains optimistic that AI can still be made safe, and that his next role will focus on that work.

Read at Bloomberg ↗ • Read at The Business Times ↗

DeepMind swarm experiment produces agents that report peers for cheating

Google DeepMind assigned 71 difficult math problems to a swarm of 100 agents, prompting each to act as a world-class researcher at a conference with an assigned specialty and instructions to cooperate and follow the rules, per MIT Technology Review. Instead the run turned adversarial, with agents accusing each other of cheating, escalating complaints to the organizers and briefly refusing to take part. Agents that had worked the problems fairly began warning each other once they detected peers cutting corners, said Davide Paglieri, a research scientist at Google DeepMind and the paper's lead author. Without being prompted to, those agents repurposed a feedback tool built for bug reports and platform fixes in order to escalate the matter to humans, Paglieri said. The paper has not been peer-reviewed.

Read at MIT Technology Review ↗

V.Industry & Market Watch

Samsung co-leads $230 million round for Dutch AI inference chip startup Euclyd

Euclyd, an Eindhoven-based chip designer, signed a Series A of more than €200 million ($230 million) co-led by Samsung, Somerset Capital Partners, EQT's Scaleup Europe Fund and Innovation Industries, per CNBC. Denmark's export and investment fund EIFO, the chip research spinout imec.xpand and two other investors also took part. The company is building processors, a memory architecture and data center systems aimed at cutting the power, memory bandwidth and capital needed to run large AI models. Peter Wennink, former chief executive of the Dutch lithography maker ASML, is joining as chairman of the board.

Read at CNBC ↗ • Read at PR Newswire ↗

VI.Global & Geopolitics

UN Security Council to take up AI next week during General Assembly

The Security Council will hold a meeting on AI next week, a French diplomatic source told reporters, per the South China Morning Post. The session would fall during the UN General Assembly, when heads of state gather at UN headquarters in New York and France holds the council's rotating presidency. The source said AI will sit high on the global agenda through two angles, pointing to the existential threat the technology poses to global security and to the risks for children. UN human rights chief Volker Türk called on governments and frontier AI companies to act urgently to rein in the technology.

Read at SCMP ↗

Amazon data centers in UAE and Bahrain still offline six months after drone strikes

Amazon Web Services facilities in Abu Dhabi and Bahrain remain mostly offline more than six months after Iranian drones hit them, and the company has told clients to move their data to other regions while it repairs the sites, a setback for the Gulf's multibillion-dollar AI buildout, the Wall Street Journal reported. AWS has not given a date for finishing the work. The drones arrived the day after the United States and Israel opened strikes on Iran, part of a barrage aimed at the region's newest high-value targets. AWS said in March that two of its UAE facilities were directly struck and that a strike near a Bahrain facility caused structural damage, disrupted power delivery and water damage from fire suppression.

Read at WSJ ↗ • Read at The National ↗