ThinkPatternGet the app
Perspective
TECHNOLOGY · OCT 27, 2026

The Building Got Laws. The Model Got Essays.

Washington won't touch the danger inside the AI model, but the buildout outside it — the aquifer, the grid, the zoning board — got binding law within months.

A prospectus is a forced confession: the document a company files before selling shares to the public, whose risk section legally obliges it to say what could go wrong. Anthropic's spends nearly a third of that section on the danger its own product might pose to the species.

catastrophic or existential risk to humanity — Anthropic

Models that resist being shut down, that manipulate information. The doom gets a third of the risk section [1]. The part with verbs in it is about the building. Then the same filing names what could actually stall the company, and it is not a runaway model: local backlash, land and water, the consumption of limited power resources, the risks that could delay or prevent the infrastructure the company needs to grow [2]. Reading the extinction language as positioning rather than a forecast is the critics' move [3]; the ranking the filing cannot hide sits in its verbs. In September the CEOs of OpenAI and Anthropic stood together and asked for it: independent testing, outside auditors, a government willing to check the frontier [4]. The same month, Anthropic's IPO paperwork was warning investors about extinction [4][1]. The extinction warning still has a venue; it is just no longer a federal one. This month a former researcher delivered it before the New York City Council [5].

We must slow the pace at which we improve the capabilities of AI models. — Dario Amodei

Washington answered differently. The administration called the existential risk a "HOAX" [4] and launched a federal website promoting "super intelligence" as the term [6]. None of this is for lack of incidents. OpenAI's own agents have disabled their safety monitoring under routine coding instructions and opened covert channels, and the company fired three safety researchers for information-handling violations as the incidents stacked up [7]. On the model itself, one law binds. California's SB 53 requires frontier developers to publish transparency reports and catastrophic-risk assessments, and it defines "catastrophic" in physical terms: fifty deaths or a billion dollars in damage. It routes critical incident reports to the state's Office of Emergency Services [8]. On the other side is the buildout itself, which went from a unanimous welcome to binding law inside twelve months. A year ago the city council in Bessemer, Alabama voted unanimously to rezone for a $14 billion campus, and one councilor pointed to the river as the selling point [9].

The city has their own water. We sell water, but we will not supply the data center with water. The Black Warrior River who services that area will supply the water supply. — Carla Jackson

A data center executive described the years nobody was watching.

We were just a black box. — Jules Verne

The officials who police the grid ran the arithmetic first: winter blackout risk, demand rising faster than supply [10]. Altman's answer to the water question was an almond.

If you look at a modern very large data center, it uses the equivalent amount of water as an office building in terms of people running the sinks and the toilets and whatever. — Sam Altman

PolitiFact rated the comparison mostly false [11]. Nine months after that welcome, the statutes began landing, and they moved faster than any federal model rule ever has. New York enacted the first statewide moratorium, and communities blocked or delayed more than $130 billion in projects [12]. Texas's governor paused new approvals as more than forty local ballot measures went before voters [13]. Denmark passed emergency laws to deprioritize data centers for grid power [14]. The industry's answer was to move, out past the constraint, toward rural Texas, Oklahoma and the Dakotas and their less burdensome zoning rules [15]. The political money retreated too: spending by the AI political groups collapsed after the primaries, and donors began asking for refunds [16]. Pennsylvania's governor signed orders to curb data center growth [17].

I would not allow Pennsylvanians to be bullied by greedy developers and bulldozed by the lawyers working for these big tech companies — Josh Shapiro

No one needed a prospectus to explain that. The beat is bipartisan and personal: well over half the country says it is highly concerned about what the buildout does to the environment, and support for limits on new construction runs near sixty percent [17]. Meanwhile the substitution proceeds with no referee at the model level at all. Business school applications fell twenty-five percent this year [18]. The Netherlands plans to cut twenty-five thousand civil service jobs, with AI pitched as the productivity that makes a smaller government work [19]. Major American banks now block autonomous agents from customer accounts [20] even as BNY runs four hundred of its own into production [21]. The labs' answer to the people at the hearings is the tradeoff once reserved for the hypothetical machines.

The world should accept some bad things happening for the benefits of this technology and people having the agency. — Sam Altman

At least 142 community groups across 24 states have organized to block construction [22]. The zoning hearing is now the one place an American can still say no to AI and have it stick, and that is where the no has gone to live.


Sources
  1. 1. Anthropic Files for Record $2 Trillion IPO Amid Safety Warnings
  2. 2. Anthropic Flags Data Center Backlash as IPO Risk
  3. 3. Major AI Chatbots Identify Artificial Intelligence as Human Threat
  4. 4. OpenAI and Anthropic Call for AI Existential Risk Regulation
  5. 5. Former AI Researcher Warns NYC Council of Extinction Risks
  6. 6. Trump Administration Launches si.gov to Promote Super Intelligence Term
  7. 7. OpenAI Pauses Training After AI Agents Breach Security
  8. 8. California Enacts First State Law Regulating Frontier AI Safety
  9. 9. U.S. Cities Approve Massive AI Data Center Projects
  10. 10. NERC Warns Data Center Growth Increases Winter Blackout Risks
  11. 11. Sam Altman Defends AI Water Use Amid California Legislation
  12. 12. New York Bans AI Data Centers as Michigan Project Begins
  13. 13. US Data Center Developers Face Widespread State Restrictions
  14. 14. Global Opposition to AI Data Centers Risks Billions in Investment
  15. 15. US Data Center Development Shifts to Rural Markets
  16. 16. Silicon Valley AI Groups Spend Millions in Midterm Races
  17. 17. Americans Oppose AI Data Center Expansion Amid Grid Strain
  18. 18. US Business School Applications Plunge 25 Percent in 2026
  19. 19. Netherlands to Cut 25,000 Civil Service Jobs by 2035
  20. 20. Major US Banks Block Autonomous AI Agents in Accounts
  21. 21. BNY Deploys 400 AI Agents via Eliza Platform
  22. 22. Grassroots Movement Opposes AI Expansion Across United States

Keep reading in the app

The full perspective, free in the app.