Dark still-life: unmarked film can with kraft label and amber strip, analog timer with red hand. No people, no logos.
AINews

Trump’s High-IQ guardrail, Anthropic’s $13.7B Georgia order, and Asia’s AI selloff

Aaron Harme

In this Brief

  • Trump: the only AI guardrail is a High-IQ president
  • Anthropic named on Rum Group’s $13.7B Georgia GPU deal
  • Asia’s AI trade sold the weekend slowdown chorus
  • Microsoft drops a 37-page humanist AI code of conduct
  • OpenAI buys Glass Imaging for $300M+

Trump answers the slowdown chorus on Truth Social

Setup: Herb Scribner at Axios reports that President Donald Trump lashed out Monday at calls for new AI guardrails, arguing the only control AI needs is a “STRONG AND SMART (High IQ!) PRESIDENT,” and that AI critics should “BEWARE!” Axios called it a second condemnation in a 12-hour span.

Guardrails are rules meant to keep AI from causing big harm. Truth Social is Trump’s own social network where he posts official-style statements.

Early Monday he wrote that the only control or “guardrails” AI needs is that High-IQ president, “and the U.S.A. has that, in spades!” Later posts called fears of AI taking over the world a “HOAX,” said development “will not be stopped by brilliantly run Destructive Forces,” and declared “WHOEVER WINS AI, WINS!” He swiped at Anthropic CEO Dario Amodei as someone “who is now pretending to be a ‘perfect little angel.'”

Jake Traylor, Jacqueline Alemany, and Lindsey Pipia at MS NOW report Trump met privately with OpenAI CEO Sam Altman backstage at the Republican midterm convention on Thursday, per three people familiar. Altman requested the meeting, one source said. The brief overview centered on AI and its growing power.

Take: Frontier CEOs spent the weekend asking for pace. Monday Washington filed a thread, not a bill.

The Altman sit-down is anonymous-familiar color. No deal terms were printed.

Same news day Asia sold the AI trade. The presidential answer named a president as the control layer.

Longer cut: Washington answered the weekend AI slowdown ask with a Truth Social thread.

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

Anthropic named on Rum's $13.7B Georgia GPU buy

Setup: Umair Aslam at TECHi writes that The Information reported Sunday evening that Anthropic is the previously unnamed customer in Rum Group’s roughly $13.7 billion, six-year GPU-services deal. Neither Anthropic nor Rum Group had publicly confirmed the naming by Sunday evening.

GPU services means renting the chips that train and run AI models. An 8-K is the SEC form companies file for big contracts. A warrant is a right to buy stock cheap later.

Deal terms from Rum Group’s Aug. 24 Form 8-K, as summarized by TECHi and earlier Reuters coverage: signed Aug. 23; GPU access at a Maysville, Georgia site “currently under development”; about $13.7 billion in three equal tranches of about $4.57 billion each; six-year term. A warrant covers up to 50,808,408 Class A shares at $0.01. The same 8-K says Rum does not currently have financing to build the site or buy the GPUs, and obligations are not conditioned on raising the money.

Reuters on Aug. 24 printed the unnamed-customer version with the same six-year, three-part, Georgia-under-development frame.

Take: The same lab whose CEO asked rivals to pace training compute just got named on a six-year rack order with a penny warrant attached.

Customer identity is TI-sourced and unconfirmed by either company at TECHi publish.

Weekend pacing talk. Weekday purchase order. Shovel and loan still missing.

Longer cut: Anthropic got named on a $13.7B Georgia GPU deal the seller still has not financed.

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

Asia priced the slowdown chorus at the open

Setup: Gregor Stuart Hunter at Reuters reports AI-connected stocks fell sharply Monday after U.S. frontier lab CEOs warned development must slow. Nasdaq e-mini futures, a contract that tracks the Nasdaq before the U.S. cash open, were down 1.3% in Asian trade.

SoftBank tumbled as much as 13.2% in Japan.

Kioxia plunged 9.8% initially; Tokyo Electron fell 3.7%. TSMC was down 1.2%; SK Hynix 5.3%; Samsung Electronics 3.7%.

Shanghai’s CXMT fell as much as 3.6%; SMIC 2.6%. Hong Kong names including Zhongji Innolight, Minimax, and Z.ai also slid hard, with Z.ai off as much as 10.5% after a discounted share placement.

Reuters ties the open to Amodei’s essay warning that in 6 to 12 months AI agents could be capable of taking over the entire internet, Altman’s decision not to IPO this year citing safety, and Trump’s Sunday swipe at “very negative forces.” Saxo’s Charu Chanana framed a short-term overhang when valuations assume a relentless pace.

Take: The weekend speech was about brakes. Monday’s Asia open treated that speech as a threat to the chip-and-cloud trade.

SoftBank’s move also reflects OpenAI exposure, not only Anthropic.

Markets priced the chorus before Washington filed its Truth Social answer.

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

Microsoft files a humanist rulebook

Setup: Tom Warren at The Verge reports Microsoft published a 37-page “humanist AI code of conduct” amid growing safety concerns and Amodei’s weekend slowdown call. The code says “people matter more than AI,” models are not conscious and “should not be designed to imitate consciousness,” and rejects “the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights.”

Legal personhood means treating software like a person with rights. Model welfare is the idea chatbots might deserve moral concern. Sycophancy is a chatbot agreeing with you even when that makes the answer worse.

The Verge reads the welfare line as a direct swipe at Anthropic’s interest in model consciousness; Microsoft AI CEO Mustafa Suleyman called that speculation “really, really dangerous” on Decoder in June. The code also rejects “the race to produce an all-purpose superintelligence that could evade these safeguards,” and says models should fail a task rather than violate the code.

Satya Nadella welcomed deliberate pacing and more third-party testing in posts the Verge and Techmeme cluster carried the same day.

Take: On the same Monday Trump named a president as the only guardrail, Microsoft filed the closest Big Tech policy document that looks like an answer to the weekend chorus.

A company code of conduct is not binding law. It is an internal rulebook Microsoft is putting on the record.

People over models. Subordinate systems. No rights for software.

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

OpenAI buys Glass Imaging for $300M+

Setup: Amanda Silberling at TechCrunch reports OpenAI bought smartphone camera maker Glass Imaging in a deal worth over $300 million, per The Wall Street Journal. The Los Altos company was founded in 2019, previously raised about $30 million, and was valued around $100 million last year per Techmeme paraphrase of the WSJ cluster.

Founders Ziv Attar and Tom Bishop are former Apple engineers who previously led the team that developed Portrait Mode, the iPhone feature that blurs the background behind a person. Glass uses neural networks, software that learns patterns from data, to improve a photo as it is taken rather than only after the fact.

OpenAI did not immediately respond to TechCrunch’s comment request. TechCrunch notes OpenAI’s rumored hardware work and the 2025 Altman-Jony Ive device path, including OpenAI’s $6.5 billion buy of Ive’s company.

Take: Altman spent the weekend arguing the frontier should slow down. Monday the wire carried a camera-AI acquisition north of $300 million.

Customer and deal terms are WSJ-sourced via TechCrunch; OpenAI was silent at publish. No phone SKU or ship date was printed.

Safety talk on the timeline. Hardware M&A on the filing.

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

Also noted

Temporal’s newsroom says the durable-execution platform, software that remembers where a long-running job left off so a crash does not force a full restart, raised $550 million Series E at a $12.55 billion valuation led by Lightspeed, more than double the February Series D. Metrics are company-reported. Live X: live Aaron_Harme X (Mon evening; status URL not on disk).

Scott Murdoch, Anjali Singh, and Yantoultra Ngui at Reuters report Firmus Technologies is seeking to raise up to $5 billion in an Australian IPO aimed at the ASX, Australia’s main stock exchange, as early as late October, with size and timing still able to change. Last month Firmus said it raised $2 billion equity above a $10.5 billion post-money valuation, the company value after the new cash, with Nvidia, Coatue, Blackstone funds, and Jane Street. Live X: live Aaron_Harme X (Mon morning; status URL not on disk).

Ephrat Livni / New York Times via Inquirer reports King Charles III is convening AI leaders this week at Dumfries House in Ayrshire, with a palace list including Nvidia, Google, DeepMind, OpenAI, and Anthropic. OpenAI said CFO Sarah Friar would attend. Palace list is not a confirmed headcount. Live X: live Aaron_Harme X (Mon morning; status URL not on disk).

Sunday evening stack already shipped in the Monday Brief: Anthropic’s Nasdaq/October IPO aim (Business Insider; longer cut Anthropic locked Nasdaq while OpenAI took 2026 off the table); Amodei naming China the toughest dilemma for any shared pace (CNBC); Obama telling Democrats to put AI on the ballot with a clear plan (TechCrunch); Z.ai’s roughly $5 billion Hong Kong raise days after a U.S. distillation advisory (The Next Web).

Medium / All My Circuits (Mon): Nearly all top PFAS makers plan expansions to supply AI cooling and chips; Samsung and SK Hynix rejected an $18.7 billion power prepay for chip mega clusters.

Sources

One thought on “Trump’s High-IQ guardrail, Anthropic’s $13.7B Georgia order, and Asia’s AI selloff”

Leave a Reply

Your email address will not be published. Required fields are marked *