Trump rejects AI guardrails; for Anthropic, $13.7 billion buys Georgia compute

Good morning.

Washington keeps handing AI safety to someone else.

President Trump posted about AI six times Monday. He said AI needs just one guardrail, a “STRONG AND SMART (High IQ!) PRESIDENT.”

The Information identifies Anthropic as the customer behind Rum Group’s six-year, $13.7 billion compute agreement. The Maysville data center has 120 megawatts of grid power and room for 180 megawatts.

Mike Johnson proposed a White House meeting with seven or eight AI leaders Sunday. Asked about a kill switch, a federal regulator or approvals before release, he said, “Maybe all of it. And those are great ideas.”

Stay curious,

Marcus Schuler

President Trump said on Monday, September 14, that AI needs only a “STRONG AND SMART (High IQ!) PRESIDENT.”

He called fears of AI destroying humanity a “HOAX.” He said the government already has regulatory power over AI companies. He also criticized Google’s €13 billion infrastructure commitment in Finland.

Dario Amodei’s Saturday essay called for slower capability gains. Elon Musk and Sam Altman supported the proposal. The Pentagon designated Anthropic a supply-chain risk earlier in 2026. Anthropic sued.

Why This Matters:

• Anthropic could face renewed federal pressure if the administration treats presidential intervention as the main safeguard for frontier AI.
• The September 24 Xi visit could provide a forum for guardrails if AI enters the state-visit agenda.

The Information identified Anthropic as the customer in Rum Group’s six-year, $13.7 billion compute contract.

Neither company has confirmed the agreement. Rum Group’s Maysville site has 120 megawatts of grid power, with expansion space to 180 megawatts. The one-cent option on up to 51 million shares could tie more Anthropic compute purchases to a larger equity position.

Fortune’s September 4 piece examines labor’s record-low share of U.S. income before an AI boom.

Labor compensation fell to 52.8% of U.S. income in the second quarter of 2026. Net U.S. imports of large computers reached a $450 billion annualized pace in August 2026.

• OpenAI on Monday, September 14, adopted safety cases before frontier reinforcement-learning runs expected to increase capability, without waiting for legislation or an antitrust waiver.
• Anthropic expects positive adjusted operating income for the quarter ending September 30, excluding stock compensation, with gross margins above 80% before partner revenue shares and model-training costs.
• OpenAI bought Glass Imaging for more than $300 million on September 14, without disclosing plans for the smartphone-camera startup founded in 2019.
• Temporal raised $550 million at a $12.55 billion valuation on September 14, after its February 2026 Series D valued it at $5 billion.
• Z.ai raised about $5 billion on September 13, including about $2 billion from a Hong Kong placement and about $3 billion from September 2027 convertible bonds.

Turn a new rule into an action map. A policy notice can bury the one sentence that changes your work. Convert the source text into an owned compliance plan before the internal briefing.

Your raw input: Paste the final rule or policy, its publication and effective dates, enforcement language, cited exceptions, the business units involved, and brief notes on your current process.

The prompt:

Why this works: Requiring a source quote keeps the model tied to the actual rule. Ownership and evidence fields convert legal language into work that can be assigned and checked.

What to use: ChatGPT handles long policy text well. Claude is a strong fallback.

Papers AI is a browser workspace from the Overleaf team that analyzes project files and turns findings into a draft. For a quick CSV brief, it keeps the source data beside the resulting note, an advantage over copying an answer from a general chat into a separate document; its Free plan costs £0, and Plus costs £10 per month as listed on September 1, 2026.

How to use it:

1. Open Papers AI in Chrome or another Chromium-based browser, click New project, and name the project.
2. Choose Local folder to keep the project on your device, then click Create project.
3. On Add files, switch to Choose from device and select a .csv file; click Add files.
4. Open the right sidebar, select AI, attach the CSV, and confirm its filename appears as a chip.
5. Ask a concrete question such as “Which category changed most, and what date drove it?” Then review the figures.
6. Add a file called notes.md in the file explorer and record the verified finding; confirm Saved in the header.

Where it falls short: Papers AI can store Excel and PowerPoint files, but it cannot open or edit .xlsx, .xls, .pptx, .ppt, or .key files.

On Sunday, September 13, 2026, House Speaker Mike Johnson rejected calls for an emergency session and an AI moratorium, citing the race with China. He called a kill switch, a federal AI regulator and mandatory government approval of frontier models “great ideas,” then named none for a vote (Implicator, September 14, 2026).

Our take: Johnson has devised a tidy AI policy. He called three safeguards “great ideas” and offered none a vote. Congress is “less qualified” than the companies it would regulate, he said, though those companies have “no consensus.” His remedy is seven or eight industry leaders at the White House.

“I’d do it tomorrow,” Johnson said of the meeting. The House is scheduled to go home until after the November elections, despite Democrats’ request to stay and advance bipartisan safeguards. Washington has found its AI kill switch: Mike Johnson’s calendar.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai