Polls | Gaza | Fact Checker | Support Us
Sign In | Join Free |
UK FACT CHECK POLITICS
UK FACT CHECK POLITICS
 Voting intention Labour Lab 27% Reform Ref 24% Con Con 18% Green Grn 11% Lib Dem LD 10% Poll tracker

Menu

Topics
About
Get Involved
Follow Us

Account

Behind the Curtain: It's not too late

axios.com 22 September 2026 at 05:12 View original article →

The original link is the material being checked. It may contain false or misleading claims — we link for transparency, not endorsement.

68/100
Verdict Mixed Mostly Reliable Opinion Commentary — factual anchors largely check out; several interpretive and unnamed-source...

The item is an Axios 'Behind the Curtain' opinion column by co-founder Jim VandeHei, published 12 September 2026 under the headline 'It's not too late' (https://www.axios.com/2026/09/12/ai-regulation-safety-plan-2026).

✓ 9 checks out ✕ 3 disputed ? 8 unverified
☰ Full analysis The complete summary ⌄

The item is an Axios 'Behind the Curtain' opinion column by co-founder Jim VandeHei, published 12 September 2026 under the headline 'It's not too late' (https://www.axios.com/2026/09/12/ai-regulation-safety-plan-2026). It is explicitly prescriptive rather than reportorial: the author states he is 'channeling' the private views of lawmakers, White House officials and AI lab staff, which makes most of its load-bearing sourcing unverifiable by design. The verifiable factual scaffolding is sound for the period: record AI infrastructure capital expenditure, a well-documented public backlash against data centres, rising Pew-measured public unease, President Trump's vocal hostility to AI guardrails, and frontier labs actively discussing an industry safety body. Two claims warrant caution. First, the assertion that 'if you listen closely to Vice President Vance, you can hear rising worry' sits awkwardly against Vance's public remarks in the days immediately after publication, when he called industry calls for regulation a 'Trojan horse' and said Americans should not be scared of AI. Second, 'We're winning the AI race now' is contested: multiple analysts and industry leaders assess the US–China model performance gap as very narrow or effectively closed. The referenced July Wall Street Journal essay could not be independently opened and is therefore Unverified rather than false. Readers should treat the five-point programme as advocacy, and the private-conversation attributions as unfalsifiable.

What checked out (9)
  • Provenance: the text matches the Axios 'Behind the Curtain: It's not too late' column dated 12 September 2026 (https://www.axios.com/2026/09/12/ai-regulation-safety-plan-2026), including the verbatim Wall Street Journal quotation and five-point structure; also summarised third-party at https://newsletter.ivodaalder.com/p/the-diplomatic-cable-002-avoiding.
  • 'We've never seen this much investment in one technology this fast in human history' — directionally verified: PwC projects roughly US$800bn of annual data-centre capex in 2026 rising to US$1.8tn a year, US$31.6tn through 2050 (https://www.pwc.com/gx/en/news-room/press-releases/2026/global-investment-in-ai-infrastructure.html); hyperscaler 2026 capex plans reported at ~US$630–700bn, up from a record ~US$388bn in 2025 (https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/).
  • 'There's now a growing public cry for attention and action' — verified: Pew reported on 18 August 2026 that 52% of Americans are more concerned than excited about AI in daily life, up from 37% in 2021 (https://www.pewresearch.org/short-reads/2026/08/18/young-adults-in-the-us-are-increasingly-wary-of-ai-concerned-it-will-take-jobs/).
  • 'These execs are spooked by the public backlash' over data centres — verified: Gallup found seven in ten Americans oppose a local AI data centre, including 48% strongly opposed (https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx); Axios reported the backlash as an existential political threat (https://www.axios.com/2026/08/19/data-centers-ai-political-opinion).
  • 'President Trump opposes regulation' — verified: Trump publicly questioned why industry leaders would seek regulation and dismissed guardrail calls in September 2026 (https://www.cnbc.com/2026/09/15/trump-opposition-to-ai-rules-undercuts-industrys-calls-for-a-slowdown.html, https://www.cnn.com/2026/09/16/politics/trump-ai-denial-hoax), following his 11 December 2025 executive order preempting state AI laws (https://www.lw.com/en/insights/ai-executive-order-targets-state-laws-and-seeks-uniform-federal-standards).
  • 'Most major labs want this [an AI governing body]' — substantially verified: OpenAI, Anthropic and Google DeepMind were reported in mid-September 2026 to be coordinating on safety standards and a self-regulatory body (https://www.cnbc.com/2026/09/15/open-ai-google-anthropic-safety.html), and Anthropic's chief executive published an essay urging stronger safety standards (https://www.cnbc.com/2026/09/15/trump-opposition-to-ai-rules-undercuts-industrys-calls-for-a-slowdown.html).
  • 'Countless AI insiders are warning of catastrophic outcomes' — verified in context: former Anthropic researcher Jacob Coxon resigned on 8 September 2026 with a widely circulated warning about self-improving systems (https://www.cbsnews.com/news/anthropic-researcher-jacob-coxon-ai-warning/, https://time.com/article/2026/09/15/ai-anthropic-researcher-quits-coxon-slowdown/).
  • The '$5,000 checks' reference is real and current: Trump promised a US$5,000 dividend contingent on Republican midterm victory, drawing bipartisan pushback (https://www.cnbc.com/2026/09/10/trump-dividend-midterms-gop.html).
  • Data-centre community-benefit conditions of the kind proposed are already being legislated at state level, e.g. Michigan bills requiring community benefit agreements (https://www.multistate.us/insider/2026/7/15/michigan-data-center-legislation-targets-energy-use-and-community-benefits).
✕ Disputed claims 3 claims ⌄
  • CONTESTED (not rated false): 'If you listen closely to Vice President Vance, you can hear rising worry.' Within days of publication Vance characterised AI firms' calls for regulation as 'like a Trojan horse' and rejected slowdown calls (https://www.benzinga.com/markets/tech/26/09/61846153/jd-vance-says-if-youre-building-frankenstein-stop-vp-pushes-for-ai-growth-not-fearmongering, https://americanbazaaronline.com/2026/09/15/jd-vance-rejects-calls-to-slow-ai-development-488157/), and told reporters Americans should not be scared of AI (https://www.pbs.org/newshour/politics/watch-vance-says-americans-should-not-be-scared-of-ai-as-calls-for-limits-grow). As the claim concerns private tone it is interpretive, so downgraded rather than falsified.
  • CONTESTED (not rated false): 'We're winning the AI race now, and may never have more leverage.' Expert assessment is divided; industry leaders and analysts place Chinese models very close behind or level with US frontier models (https://www.poynter.org/fact-checking/2026/whos-winning-ai-race-us-vs-china/, https://www.csis.org/analysis/chinas-open-weight-challenge-us-ai-leadership).
  • CONTESTED (not rated false): the implication of 'no serious effort to control it'. Federal and state activity exists — a December 2025 preemption executive order, numerous AI bills in the 119th Congress, and extensive state data-centre statutes (https://www.dlapiper.com/en-us/insights/publications/2025/12/new-executive-order-aims-to-preempt-state-ai-regulation, https://www.afslaw.com/perspectives/alerts/state-regulation-data-centers-2026-shifting-landscape) — though none amounts to the frontier-safety regulator the author advocates.
? Unverified claims 8 claims ⌄
  • The July Wall Street Journal essay attributed to the author: the quoted passage appears only within the Axios column in available results; the original WSJ piece could not be opened or dated independently — Unverified, not disputed.
  • 'Based on our months of conversations with lawmakers, White House officials and AI labs' and 'almost every person we talk to privately' — unnamed, non-attributable private sourcing that is structurally unverifiable.
  • 'The chances of AI worst-case scenarios will explode, likely early next year' — a prediction about 2027; neither verifiable nor falsifiable at assessment time.
  • 'Most major labs ... would agree to [a body with] the power to review and shut them down' — lab support for a safety body is documented, but agreement to shutdown authority is not evidenced in any located source.
  • 'A broad understanding, at least in private, of what could be done ... with bipartisan and industry buy-in' — no public evidentiary trail; House Democrats' reported consideration of a select AI committee (https://www.nextgov.com/artificial-intelligence/2026/09/democrats-mull-creating-select-ai-committee-if-they-win-house/415885/) is suggestive but does not establish bipartisan buy-in.
  • 'The top AI companies would take this deal in a heartbeat' on data centres — no company has publicly committed to the described package of energy supply, water offsets, transparency and resident dividends.
  • 'The technology is vastly better' over two months and imminent 'recursive self-learning' — contested research questions; recursive self-improvement remains an area of active study rather than demonstrated deployed capability (https://iclr.cc/virtual/2026/workshop/10000796).
  • Implicit claim that China would consider a nuclear-style AI framework: a bilateral US–China AI governance channel was announced around the May 2026 Trump–Xi summit (https://www.iiss.org/online-analysis/online-analysis/2026/06/military-ai-governance-under-strain-the-uschina-dialogue/), but Beijing's acceptance of a binding regime is Unverified.
⌕ Sources & how we checked Search journal, source grades, confidence ⌄
Confidence

Medium-High (0.76) — Confidence is supported by successful provenance identification of the exact column and by multiple independent, dated, high-quality sources (Pew, Gallup, PwC, CNBC, CNN, PBS, Poynter, CSIS, IISS, Congress.gov) covering every high-priority checkable claim. It is held below high because: (1) the article's central evidentiary basis — unnamed private conversations — is unfalsifiable, so a large share of its assertions can never move beyond Unverified; (2) the self-cited Wall Street Journal essay could not be opened or independently dated; (3) the supplied text had its hyperlinks stripped, removing the author's own sourcing trail; (4) two key contested items (Vance's private sentiment, the US–China standing) are interpretive or genuinely disputed among experts rather than resolvable; and (5) the political and capability landscape is moving quickly, so assessments dated within days of publication may shift again.

Search journal

Axios Behind the Curtain VandeHei AI "all-hands-on-deck" AI governing body column

VandeHei Wall Street Journal essay July 2025 AI race "we're blowing it"

Trump executive order state AI laws preemption regulation 2025

JD Vance AI concerns worry comments 2025

Vance AI comments September 2026 worry regulation

poll public opinion AI regulation data centers backlash 2026

Pew Research Center AI concerns poll August 2026 52% more concerned

AI capital expenditure record investment data centers 2026 trillion

AI labs call for federal regulation OpenAI Anthropic 2026 support oversight body

Trump statement AI regulation September 2026 "call for Regulation" industry leaders

is the US winning the AI race against China 2026 assessment leadership models

US China AI talks governance framework nuclear arms control model 2026

recursive self-improvement AI models 2026 evidence self-improving

Anthropic researcher Jacob Coxon warning AI September 2026

Congress AI select committee legislation federal AI oversight agency 2026

tariff dividend $5,000 checks Trump 2026 status

data center community benefits agreements water energy legislation states 2026

"VandeHei" opinion Wall Street Journal AI July 2026 "generational challenge"

Article metrics

Emotion 12% · Reading grade 11.4 · 804 words

≡ The article we checked Full text as retrieved ⌄
Illustration: Brendan Lynch/Axios The White House, Congress and the country's leading [AI companies]( have allowed AI to grow faster, stronger, more powerful and wildly lucrative (and dangerous), with no serious effort to control it or spread the benefits beyond the super-wealthy. * It's not too late to change this. But if they all dither, duck or dance around it, the chances of AI worst-case scenarios will explode, likely early next year, according to almost every person we talk to privately who truly understands AI. **Why it matters:** The technology is getting better, faster than anyone envisioned — with the possibility rising that AI will self-improve and do wildly unpredictable things. * **The good news:** There's now a growing public cry for attention and action and, at least in private, a broad understanding of what could be done relatively fast with bipartisan and industry buy-in. **The big picture:** In July, I wrote an [essay]( for The Wall Street Journal saying America "faces one of those rare, historic moments when government, business, schools and families could be working together to meet a truly generational challenge: winning the AI race. So far, we're blowing it." * Two short months later, the technology is vastly better — and the societal response vastly dumber. As a nation, we're choosing to [amuse or distract ourselves]( **Here's how to change that fast,**based on our months of conversations with lawmakers, White House officials and AI labs. Think of what follows as me channeling their private views: 1. **Sound an all-hands-on-deck alarm.**Every political topic is trivial and small compared to AI. We will look back in shame if we get spooled up about trans bans or $5,000 checks when countless AI insiders are warning of catastrophic outcomes. Let's say they're all paranoid and delusional. If so, it still doesn't matter: we've never seen this much investment in one technology this fast in human history. It's seeping into every industry and every job and every community and every screen, at a speed never before witnessed by man. Yes, we need a special session of Congress and select committees. But we need even more: AI execs, bioterror specialists, job-training wizards and safety experts convening to plan for scenarios and act fast. 2. **An AI governing body with teeth and speed, ASAP.** This isn't rocket science. Most major labs want this and would agree to it, as long as it applies to all rivals and competitors. This body would need the expertise to understand these models, and the power to review and shut them down if they display dangerous behavior. It can't operate at typical regulatory speed — it needs to move as fast as the technology. It's nonsense to think you can't advance AI development fast and regulate it wisely simultaneously. 3. **A universal standard for data centers to benefit communities and taxpayers.**The top AI companies would take this deal in a heartbeat: They get to build and operate data centers provided they supply a community's energy, offset water use or environmental damage, transparently report on their building and business plans, and spread the financial benefits to local residents. These execs are spooked by the public backlash and ready to aggressively spread the financial benefits, with dividends or cash. So make them. 4. **Engage China.** We shouldn't trust them, and they shouldn't trust us. But trust me: the only way to ultimately control AI on a worldwide scale in coming years will be for the two superpowers to create some kind of framework modeled after how we've safely governed nuclear weapons for decades. We need to start this discussion now so potential collaboration is possible in the coming years. To those who think China will never consider this seriously: Are you sure an authoritarian regime really wants AI once it hits [recursive self-learning]( and can't be controlled by anyone, including an autocratic regime? We're winning the AI race now, and may never have more leverage to shape this joint effort. 5. **Build coordinating infrastructure.** What I wrote in the WSJ is even more needed today: "Bring together the best and brightest drawn from the federal government, from the leading AI companies, from business and labor, from economics, from public health, from ethics to form a working group with actual authority. People who are paid to anticipate, not to react haphazardly." This group would have four primary functions: map potential problems and upsides before they hit, build the (policy) playbooks before they're needed, level with the American people along the way, and expedite work to deliver widespread AI benefits. **Reality check:**Yes, President Trump opposes regulation or bipartisan efforts. Yes, AI companies are obsessed with more money, more speed. But the politics are shifting fast: If you listen closely to Vice President Vance, you can hear rising worry. At this pace, toxic AI politics could force mind-changing that was unthinkable two months ago.

Share this fact check

← Check another article or image