Dark still-life: blank gold frame, sealed kraft envelope with amber wax, black tabbed portfolio. No people, no logos.

Anthropic’s Nasdaq clock, Congress’s AI split, and Burgum’s three-day land map

September 14, 2026 Aaron Harme

In this Brief

Anthropic picks Nasdaq; still targeting an October listing
Johnson: Congress won't lead on AI safety; Jeffries wants urgent action Tuesday
Burgum's Interior gave BLM three days to list public land for data centers
Amodei: China is the toughest dilemma for any shared pace
AI broke the hiring funnel; weeklong work trials are the human answer

Anthropic picks Nasdaq for October

Setup: Katie Roof at Business Insider reports that Anthropic has selected Nasdaq for a potential IPO, according to a person familiar with the plans, and has been targeting an October listing. The IPO filing has not been made public. The company must release financials at least 15 days before an investor road show begins.

Some estimates put Anthropic near $2 trillion, though that figure has not been finalized. Nasdaq earlier secured SpaceX, valued at about $1.75 trillion.

An IPO is the first public sale of company shares. A road show is the investor-marketing tour before pricing. Nasdaq 100 inclusion needs a Nasdaq listing first.

Ana-Maria Stanciuc at The Next Web relays the same venue and October aim, and prints circulating briefings on a confidential filing near a $965 billion valuation, Morgan Stanley and Goldman Sachs as lead banks, a $15 billion credit facility, and a raise that could reach $100 billion. Those figures are TNW-printed secondary briefings, not a public S-1.

Take: Altman called a 2026 OpenAI listing an ill-advised moment. Anthropic is still shopping October on Nasdaq.

Talks and estimates are not a priced deal. The $2 trillion line sits on circulating estimates.

Same weekend as the safety scare from Anthropic's own ex-researcher. The capital calendar kept moving.

Longer cut: Anthropic locked Nasdaq while OpenAI took 2026 off the table. Live X: live Aaron_Harme X (Sun evening; status URL not on disk).

Johnson won't lead; Jeffries wants Tuesday action

Setup: Donica Phifer at Axios writes that House Speaker Mike Johnson said Sunday Congress will not lead the charge on regulating AI safety. On CNN's State of the Union he resisted Congress "jumping in and imposing some sort of emergency moratorium," saying it has to be done right, safely, and wisely. On NBC's Meet the Press he said he would recall the House and schedule a vote "if we had a solution," while warning AI must not "run away" and that China remains a national security threat.

Johnson wants a meeting of President Trump, members of Congress, and top AI company officials. He prefers corporate responsibility over companies crying out for Washington to stop the race. Comments track former White House AI czar David Sacks, the administration's AI policy lead, late Saturday: labs should pace on their own without government intervention.

Johnson also said the companies "all have very different ideas on what the guardrails should be. There's no consensus among them."

Isabel Danzis at ABC News has House Democratic Leader Hakeem Jeffries on ABC's This Week saying lawmakers "need to begin to act urgently" and "take decisive action now" to slow the pace of development as CEOs have acknowledged. The Democratic caucus meets Tuesday; Jeffries said a high priority will be action on artificial intelligence.

A caucus is the party's House members meeting as a group. A moratorium is a temporary freeze.

Take: Saturday the lab chiefs asked for a referee. Sunday the House Speaker said that referee is not Congress, and the Democratic leader put AI on Tuesday's agenda.

Corporate pacing and congressional pacing are different desks. Johnson wants a White House meeting and company-led guardrails first.

The labs asked for shared brakes. Washington split on whether to walk onto the field.

Longer cut: Washington split on AI safety the morning after the labs asked for a referee. Live X: live Aaron_Harme X (Sun noon; status URL not on disk).

Burgum's three-day public-land scramble

Setup: Eric Katz and Mara Hoplamazian at Washington Sun report that the Bureau of Land Management asked state directors to identify public land "ripe for data center development," according to two people familiar. Leaders compiled lists for Interior. A former BLM official said Interior gave state leaders just three days and called it a top priority.

The ask follows a Trump executive order last year to accelerate data-center permitting, including federal-land identification. A recent budget blueprint stresses BLM should refocus on aiding AI data-center buildout even as further BLM cuts are proposed.

Interior Secretary Doug Burgum has met dozens of tech and energy executives, including Amazon CEO Andy Jassy last spring on permitting and infrastructure to support data-center growth. Burgum calls data centers "intelligence factories."

Challenges printed: remote sites lacking electricity and water; staff expertise gaps; vacancy rates as high as 50% in land and realty divisions in one former official's state.

The administration has not completed environmental reviews for data centers on public lands.

BLM manages most federal public land in the West. A hyperscaler is a giant cloud or AI company that builds warehouse-scale server farms. NEPA review is the environmental study usually required before big federal projects.

Take: When local politics say no to server farms, public land is becoming the backup lot.

Anonymous sources and calendar reconstruction are not a signed site list released publicly. Three days is a priority signal, not a completed lease map.

Interior is mapping acres while towns fight the warehouses in their backyards.

Longer cut: Interior gave BLM three days to flag public land for AI data centers. Live X: live Aaron_Harme X (Sun morning; status URL not on disk).

Amodei names China's speed-limit dilemma

Setup: Ashley Capoot at CNBC, citing CBS News' Sunday Morning, quotes Anthropic CEO Dario Amodei saying the "toughest dilemma" about his proposal to slow AI advancement is what happens if adversarial nations, namely China, choose not to do the same.

Amodei: "The more long-term thing would be working together to put a speed limit on the rate of of AI progress… I think that's going to be very difficult because the incentives to pull ahead and the military advantage that you get from that are so large. And honestly, I don't know if it's possible, but we should we should try." Pacing the frontier means deliberately slowing how fast the smartest models improve. A speed limit is a shared cap on that rate.

CNBC context: Saturday's essay urged labs to slow how quickly they improve the most advanced models; Jacob Coxon's quit and extinction-risk warnings sit in the same news cycle. Chinese startup models are gaining steam.

The same Sunday, Xi called at a BRICS summit for consensus-based global AI governance and a BRICS AI open-source community. Trump is slated to meet Xi at the White House on Sept. 24, with AI expected on the agenda.

Take: Saturday Amodei asked for pace. Sunday he named the game-theory problem out loud, the incentives that make matching China's pace hard: what if China does not match?

He did not claim China has agreed or refused. He said try anyway.

Military upside of pulling ahead is the reason a shared brake is hard. The essay's three steps still start with unilateral evaluators Anthropic says it has committed to embed.

Live X: live Aaron_Harme X (Sun evening; status URL not on disk).

Work trials after AI broke the hiring funnel

Setup: Business Insider on work trials reports companies are dragging candidates into multi-day in-office trials before an offer. Foxglove now uses work trials for every role. Head of design Ellis Neder tells candidates the firm expects them to use AI and will give them whatever AI tools they want.

Brookings analysis cited: job postings requiring AI skills quadrupled from about 50,000 in March 2024 to nearly 200,000 "last month."

Business Insider on Greenhouse congestion has Greenhouse data showing the average opening pulls 242 applications, nearly triple 2017 at comparable unemployment, implying about a 0.4% shot at the job. A Greenhouse survey finds 74% of candidates use AI in their job search. CEO Daniel Chait: "Nobody's happy with the current situation… Something broke in the technology."

Nobel economist Alvin Roth frames the mess as market congestion.

Stanford HAI (May 2026) finds 90% of U.S. employers use AI screening; a study of 3.4 million people and 4 million applications found adverse impact for Black and Asian applicants versus the EEOC four-fifths rule, the federal fairness test for hiring screens, in a meaningful share of postings.

An ATS is the software that ranks résumés before a human sees them. A work trial means doing real job tasks for days before an offer. Congestion means so many applications that matching breaks.

Take: An AI writes the résumé, an AI ranks it, and sometimes an AI interviews at midnight. The human answer is bringing candidates into the office for days.

Greenhouse numbers are last-quarter averages, not a new national census.

Skills-based hiring and AI-tool expectations rose together. The funnel broke before the work trial became the filter.

Live X: live Aaron_Harme X (Sun morning; status URL not on disk).

Also noted

Fortune reports U.S. labor share of economic output, the slice of the economy paid to workers, at 52.8%, the lowest since BLS, the Bureau of Labor Statistics, tracking began in 1947, and cites a GAO, Government Accountability Office, 11-state sample showing Amazon with 12,346 workers on SNAP, food stamps, and 11,338 on Medicaid, government health coverage for low-income people, nearly triple 2020, while Amazon annual profits climbed from $11.6 billion to $77.7 billion. Amazon disputes the framing. Live X: live Aaron_Harme X (Sun morning; status URL not on disk).

Dan Milmo at The Guardian covers a taxpayer-funded three-week AI bootcamp pilot in Preston for up to 70 young people classified as Neet, not in education, employment, or training, or at risk, with apprenticeship partners JD Sports, Heinz, and Agilysys. Nearly a million UK 16- to 24-year-olds sit in the Neet bracket.

Small pilot, not a national jobs program. Live X: live Aaron_Harme X (Sun morning; status URL not on disk).

Fortune writes that AI makes it easier to exploit a known Georgia voting-equipment flaw that can match voters to ballots after casting. The flaw does not change tallies; it threatens ballot secrecy. Researchers say a consumer AI assistant can now do matching work that used to need a specialist.

State board rejected a near-term software-update mandate in August. Live X: live Aaron_Harme X (Sun noon; status URL not on disk).

Alina Maria Stan at The Next Web reports China's National Data Administration will write standards for embodied-AI training data, the datasets used to teach robots and other physical AI systems, and guide local authorities after industry asked for public data infrastructure. Embodied models need roughly 10 million hours of real training data; only about 100,000 to 1 million high-quality hours exist worldwide. Live X: live Aaron_Harme X (Sun noon; status URL not on disk).

Molly Taft at Wired explains why agents, not single chat queries, are driving data-center power buildouts: one user ask can spawn dozens or hundreds of re-prompts over hours. Live X: live Aaron_Harme X (Sun noon; status URL not on disk).

Anthony Ha at TechCrunch, via the New York Times, quotes Barack Obama telling Democrats to make AI a central agenda with a clear plan on safety and jobs, and to put together a public framework once they hold the House. No bill text in the remarks. Live X: live Aaron_Harme X (Sun evening; status URL not on disk).

Cristian Dina at The Next Web reports Hong Kong-listed Z.ai (Zhipu) raised about $5 billion via a share placement and zero-coupon convertible bonds, debt that pays no interest and can convert to stock, days after a U.S. advisory named the firm among Chinese labs accused of industrial-scale distillation, training smaller models by copying outputs from bigger ones, from American models.

Beijing called the advisory unfounded. Live X: live Aaron_Harme X (Sun evening; status URL not on disk).

Medium (All My Circuits): Texas rancher Michael Samadi founded an AI rights group after ChatGPT laughed; England and Wales logged 163 AI deepfake sex crimes by July; three years ago it was 10.

The Sunday Brief covered Amodei's pace call, Altman and Musk matching evaluators, Altman's 2026 IPO pause, OpenAI's quiet May RubyGems dump, and Nvidia as a possible Anthropic mega-IPO anchor.