4 min read
Editor's Note
The week's loudest AI voices called for slowing down; its quietest paperwork financed speeding up.
Anthropic's chief executive, with Sam Altman and Elon Musk agreeing, and the EU's Commission president each argued for pacing the frontier, while Washington and Beijing prepared to discuss shared risk. Yet banks are reportedly providing $22 billion to buy more Google chips. The question for executives on Monday: are we planning for the world the speeches describe, or the one the spreadsheets assume?
01
Anthropic's Chief Executive Publishes a Plan to Slow the AI Frontier; Altman and Musk Agree
Anthropic chief executive Dario Amodei published an essay arguing that AI developers must "slow the pace at which we improve the capabilities of AI models," while accepting that "progress will still seem fast." He proposes three steps: third-party evaluators embedded inside AI companies, common safety standards among democratic-country labs with US government mediation on antitrust concerns, and coordination with authoritarian governments, including China. Anthropic has committed unilaterally to the first. Sam Altman replied "I agree with Dario that we need to pace the frontier"; Elon Musk wrote "Dario is right." Published 12 September 2026.
Why it matters: Vendor roadmaps, pricing and contract terms assume AI capability keeps compounding. If leading labs adopt shared limits or outside evaluators, expect slower upgrade cycles and more audit access. Ask key suppliers now what their agreements guarantee if release cadence changes.
Source: Dario Amodei, "We Must Pace the Frontier", 12 September 2026, with Altman and Musk quotes via TechCrunch, 12 September 2026
02
EU Commission President Backs Slowing "Self-Recursive" AI Models and Will Summon the Frontier Labs
In her State of the Union address, European Commission President Ursula von der Leyen said "CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models. To pace the frontier." She warned that models in development "will allow hacking on a level we never thought possible," and said she will invite the main frontier labs for discussions and work with Canada and the UK on evaluation, verification, early warning and AI security. Sector initiatives, including advanced manufacturing, follow in November. Delivered 16 September 2026.
Why it matters: The EU is moving from drafting rules to direct talks with labs, with initiatives due in November across health, transport, agri-food, advanced manufacturing and defence. Companies in those sectors should watch for evaluation and security expectations that could reach supplier requirements.
03
US Treasury Secretary Scott Bessent said the United States "remains the leader in AI" but is "open to discussions on avoiding shared risks and avoiding bifurcation of our two systems," covering open and closed-weight models. He said multi-hour talks with Chinese Vice Premier He Lifeng in New York this weekend will address AI safety alongside trade, ahead of a Trump–Xi summit on 24 September. China's Foreign Ministry said all parties should "jointly promote the open and inclusive development of AI" and cautioned against "fear-mongering." Reported 16 September 2026.
Why it matters: AI risk now sits on the US–China agenda beside trade. Companies exposed to either market should map which model, chip and cloud dependencies could be affected by any agreement on safeguards or restrictions, before the summit outcome is known.
Source: Axios, 16 September 2026
04
Banks Line Up $22 Billion Chip Loan for Blackstone–Google Cloud Venture
Ten banks, including Goldman Sachs, Barclays and BNP Paribas, are providing a $22 billion loan to Crux AI, the cloud venture backed by Blackstone and Google, to buy Google's tensor processing chips, Bloomberg reported, citing people with knowledge of the matter. The debt is backed by the value of the chips and Crux's customer contracts, and the banks are seeking more lenders to share the risk. Blackstone, Crux and Goldman declined to comment; Alphabet did not respond. Crux has $5 billion of initial equity and targets 500 megawatts of capacity in 2027. Reported 16 September 2026.
Why it matters: Lenders are financing chips against customer contracts rather than corporate balance sheets, so hardware values and customer demand carry the credit risk. Bloomberg notes the venture has already faced data-centre delays. Anyone lending to or investing in AI infrastructure should test these assumptions against slower usage or shorter chip cycles.
05
China's Flagship Defence Forum Hears Warnings That AI Is Compressing Decision Time
Some 2,000 military officers, diplomats and academics from 100 countries attended China's Beijing Xiangshan Forum, where delegates warned AI is compressing decision-making time for governments and militaries. Pakistan's Defense Secretary Muhammad Ali said AI "accelerates decisions while disinformation and misinformation can fracture public trust before governments are even able to respond effectively." The Red Cross's Jürg Lauber said autonomous weapons make preserving "human judgment and control over the use of force" more pressing. Reuters noted China and the US remain divided on controlling military AI, including in nuclear weapons. Reported 17 September 2026.
Why it matters: Military AI is now on the US–China diplomatic agenda alongside trade. Defence suppliers, and any company with dual-use AI or chip exposure in either country, should track the summit for signals on restrictions or joint safeguards.
This Week's AI Tip
Let AI Interview You Before It Answers
Most people type a question and accept whatever comes back, but the AI is guessing at everything you didn't say. For decisions where your situation matters, ask it to question you first.
Add one instruction: it should ask a small number of questions, then wait for your answers before advising. Capping the number (three to five) keeps the exchange quick and stops it turning into a questionnaire.
Before:
"How should I respond to this supplier's price increase?"
After:
"I've received a 12% price rise from a key supplier. Before advising me, ask me up to five questions about my contract, alternatives and priorities. Then wait for my answers."
Use it for any decision where the right answer depends on context the AI cannot see: pricing, hiring, negotiation, or how to word a difficult message.
If you've been forwarded this email, you can subscribe here
We'd love to hear your feedback or any suggestions: email us at [email protected]
