Dark still-life: brass desk timer with cyan LED tip, sealed folders, face-down phone. No people, no logos.

Huang’s All-In no-brake call, three-lab safety talks, and Dreamforce’s no-new-laws line

September 16, 2026 Aaron Harme

In this Brief

Huang told Trump on All-In: no industry AI slowdown
OpenAI confirmed weeks of three-lab safety talks
At Dreamforce, Huang said the industry needs no new AI laws
EPA repeal keeps old coal and gas viable for AI power
Commerce vs Kalshi on AI-compute price curves

Huang answered a live White House call with applause for no brake

Setup: Herb Scribner at Axios reports that Nvidia CEO Jensen Huang took a surprise call from President Donald Trump onstage Monday at the All-In Summit in Los Angeles and put the president on speaker. Trump called AI-takeover fears a “hoax,” blamed “political people” and China, and said the robots would not take over.

An All-In Summit is a big tech and investor conference tied to the All-In podcast. A frontier slowdown means labs agreeing to improve the most advanced AI models more slowly.

Amanda Silberling at TechCrunch reports Huang answered, “You’re right. We’re not going to let that happen, sir,” as the crowd applauded. Trump also said the industry should move “prudently” without stopping. Separately onstage, Axios notes Huang praised Anthropic whistleblower Jacob Coxon, rejected end-of-world predictions as “not grounded in science,” and said companies should pause if they feel out of control.

Take: Monday’s Truth Social thread named a High-IQ president as the only guardrail. Monday’s stage call put the chipmaker on the line live.

Huang’s commercial interest is unsold chips under an industry-wide brake. The room applauded the no-slowdown line.

Company-level pacing stayed on the table. An industry brake did not.

Longer cut: The chipmaker answered a live White House call with applause for no industry brake.

The public slowdown pitch landed after private three-lab talks

Setup: Rebecca Bellan at TechCrunch reports that OpenAI global policy chief Chris Lehane told reporters Tuesday the company has been working with Anthropic and Google DeepMind on AI safety for weeks, as first reported by Bloomberg. Lehane was in Washington on catastrophic AI risks.

An antitrust waiver is special legal permission so rivals can coordinate without being accused of collusion. An independent verification organization is an outside group allowed in to check safety work.

TechCrunch says some executives, including Sam Altman, had flagged that coordination could raise antitrust risk if it suppressed competition, and that Dario Amodei’s weekend essay floated a narrow government waiver. Lehane reportedly said the firms do not need one. The Information, via TechCrunch, reported the three companies working on a standards body; Altman reportedly told staff that would need to happen without U.S. government support.

Brendan Bordelon at Politico reports OpenAI backs a FRONTIER Act provision, a House bill on frontier-lab safety, from Reps. Jay Obernolte and Lori Trahan that would require top labs meeting high revenue and compute thresholds to allow those independent verification organizations in. Lehane told a sponsor OpenAI can support that IVO piece.

Take: The weekend looked like rival CEOs suddenly agreeing to slow down. Tuesday’s reporting says the private coordination was already running.

No deal terms from the talks were printed. No waiver ask from OpenAI’s lobbyist.

Public chorus first on the timeline. Multi-week three-lab work confirmed after.

Longer cut: The public slowdown pitch landed after weeks of private three-lab safety talks.

At Dreamforce, Huang said no new AI laws

Setup: Ashley Capoot at CNBC reports that Anthropic CEO Dario Amodei and Nvidia CEO Jensen Huang shared differing views on AI safety during Salesforce CEO Marc Benioff’s Dreamforce keynote Tuesday in San Francisco, with an audience of about 12,000 at Moscone Center.

Dreamforce is Salesforce’s big annual customer conference. A false choice here means claiming you must pick only speed or only safety.

Amodei, days after his weekend essay urging the industry to slow model-development pace, said leading means setting an example so “everyone can always be better.” Huang told Benioff the industry does not need new laws or regulations, that market forces already exist, and that “It’s a false choice. You could definitely have both at the same time.” He added: run as fast as you can, but pause if the company feels out of control or the product will not be safe.

CNBC notes the same Salesforce day brought a new reasoning model for Agentforce, Salesforce’s AI agent product line, on Nvidia’s Nemotron, open-weight models anyone can download and run, and that Salesforce and Anthropic expanded a “Claudeforce” partnership late last month, Anthropic models wired into Salesforce tools. Benioff told reporters labs should slow or speed as they see fit, then be held accountable.

Take: One day after applauding Trump’s no-brake line on All-In, Huang told a packed Moscone hall the statute book does not need a new AI chapter.

Amodei got the example-setting frame on the same keynote. Huang got the no-new-laws line.

Self-pace stayed. Industry-wide statute did not.

Longer cut: Huang told 12,000 people at Dreamforce the industry does not need new AI laws.

EPA repeal keeps old coal online for the AI power crunch

Setup: Matthew Daly at AP reports the EPA said Monday it is repealing rules that limit greenhouse gas emissions from coal- and natural-gas power plants and taking a separate related step aimed at blocking future administrations from regulating that climate pollution. Greenhouse gases are gases that trap heat in the atmosphere. The Federal Register is the official U.S. rulebook where final regs publish.

EPA said revoking the Biden-era power plant rule would save industry more than $300 billion. Assistant Administrator Aaron Szabo said utilities can decide for ratepayers on cost instead of being required to shut facilities.

The rule takes effect soon after Federal Register publication; legal challenges are expected.

A second proposal, expected final next year, follows EPA’s earlier move against the scientific finding that underpinned U.S. greenhouse-gas regulation.

Announcements landed as EPA chief Lee Zeldin, Interior’s Doug Burgum, and Energy’s Chris Wright hosted a G20 energy ministerial in Houston.

Fortune reports the biggest immediate impact experts flagged is keeping more old coal and gas plants viable longer to help power data centers and the AI boom. A data center is a warehouse of computers that train and run AI.

Opportune’s Dan Romito called coal a “Band-Aid” stopgap for three, four, maybe five years while geothermal and nuclear catch up.

Zeldin called it the largest power-sector deregulatory action ever. Biden’s 2024 standards had aimed at about 90% capture or cut of climate pollution.

Take: The same week frontier CEOs argued about slowing training, Washington made it easier to keep yesterday’s plants plugged in for tomorrow’s racks.

EPA asserted power-plant emissions have no material climate impact; critics called that a lie. Fortune’s stopgap frame is expert color, not EPA text.

Safety chorus on the labs. Coal as a bridge for the load.

Commerce vs Kalshi on AI-compute price curves

Setup: Semafor reports sources saying the Commerce Department last month ordered Kalshi to take down a product tracking the price of AI compute, citing national security, and that Kalshi quietly complied while many underlying markets remain.

A forward curve is a chart of what the market implies a price will be later. AI compute here means renting time on the chips that train and run models. The CFTC is the U.S. regulator for futures and many prediction markets.

Semafor says Commerce also pushed the CFTC to pause approval of new compute contracts for 60 days. Kalshi declined to comment. A Commerce spokesperson said the department “has never once asked Kalshi to take down this market or any other markets” and called the story false.

The CFTC did not respond.

Market-participant color in Semafor, not confirmed, floated concern that thinly traded compute markets could be manipulated to show a sharp drop in older-chip costs and rattle AI stocks and debt.

Gizmodo reports the curves launched in July for Nvidia B200, H200, and A100 rentals as benchmarks for the implied future price of one GPU-hour; the curves themselves were not tradable. Kalshi CEO Tarek Mansour said in the July release that compute is the new oil and Kalshi aims to be where buyers and sellers manage that risk. Gizmodo notes the CFTC opened a 60-day public comment period on compute derivatives in August.

Take: Labs argued all week about slowing models. Washington tripped over a different throttle: whether GPU-hour prices get a public futures-style signal at all.

Semafor’s order is sourced and disputed. Commerce’s denial is on the record.

Chip rent as a chart. National-security claim versus “this story is false.”

Live X: Aaron Harme on Polymarket / CFTC.

Also noted

Britney Nguyen / Dow Jones Newswires via Morningstar reports cybersecurity names led the S&P 500 Monday amid AI fears: CrowdStrike +13.8% to an all-time closing high, Palo Alto Networks +13%, Fortinet +9%. The iShares Expanded Tech-Software Sector ETF (IGV) beat the iShares Semiconductor ETF (SOXX) by 10.67 percentage points Monday, the largest single-day gap on record per Dow Jones Market Data. A semiconductor ETF is a fund of chipmakers; cybersecurity stocks sell tools to stop hacks.

Hugh Langley at Business Insider reports Google opened Anthropic’s Opus 5 coding model to engineers across the company inside Antigravity, its internal developer platform, with per-user quotas, per two employees and a spokesperson. Gemini remains primary; third-party models are supplementary. Google has announced plans to invest up to $40 billion in Anthropic. A coding model helps write and fix software; a quota caps how much each person can use.

Business Insider and The Independent report former Google DeepMind AGI-safety researcher Bilal Chughtai resigned, saying AI “has the potential to kill us all” and that alignment understanding is “extremely rudimentary” versus rising capabilities. He left in July 2026 after about a year and a half in London and joined BlueDot Impact, which said it recently raised $25 million. Google did not comment. Alignment means making sure a system does what people intend.

Julie Bort at TechCrunch reports Artificial Intelligence Underwriting Company (AIUC), from early Anthropic hire Rune Kvist and former COO of METR, an independent AI-safety evaluator, Rajiv Dattani, raised a $40 million Series A, an early major venture round after seed, led by Ribbit Capital ($55 million total with prior seed). The product is a SOC 2-style third-party audit for AI agents (~5,000 tests, ~100-page report) with named customers Cursor, Lovable, Harvey, and ElevenLabs. Underwriting here means an independent check on where an agent is safe to buy; SOC 2 is a common software security audit.

Annie Palmer at CNBC reports Elon Musk at Monday’s All-In Summit floated a rival-run “test harness” so xAI, OpenAI, Anthropic, Google, Meta, and three or four leading Chinese companies let competitors grade models before public release. A test harness is a shared set of safety checks one lab can run on another’s model. Musk said rivals have not agreed; he thinks China might.

Google’s Gemini Audio Team launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking for real-time voice dialogue that can keep talking while deeper reasoning runs in the background, with visual context and mid-conversation switches across 97 languages. Extended Thinking printed #1 on Artificial Analysis’ Speech to Speech Quality Index at 82.6, plus 68.6% and 35.1% on named voice-agent benchmarks and 97.7% on Big Bench Audio. Available via Gemini API, Workspace, and the Gemini app.

CNBC reports the Senate blocked the Clarity Act on a 50–49 cloture vote, short of the 60 needed to start debate. The bill would split crypto oversight between the SEC and CFTC. Cloture is the 60-vote gate to begin Senate debate. Midterms are about seven weeks out.

Prior Tuesday longer cuts already live: Huang All-In; three-lab talks; Dreamforce. Optional prior roundup: Tuesday morning Brief.