The Totalitarian Program of the AI Doomers

Abstract

Yuddites – people who believe artificial superintelligence will kill everyone – want to create a global totalitarian government with the power to (among other things):

• impose an indefinite worldwide ban on frontier AI models (anything more powerful than GPT-4)
• force chip manufacturers to install
  • mandatory hardware-level surveillance software
  • remote kill switches
  • compute caps
• force all frontier AI developers to be licensed
• impose 20-year prison sentences on frontier AI developers
• launch rocket attacks at non-obedient data centers

This post documents their aspirations, what they have already achieved, and what you can do to stop them.

What Do Doomers Want?

These are not hypotheticals – several already exist as draft treaty text or as introduced legislation.

1. Enforce an indefinite, global ban on frontier AI models (systems more powerful than GPT-4).[1] Yudkowsky argued the moratorium on new large training runs must be “indefinite and worldwide,” with all large GPU clusters shut down and a ceiling placed on training compute. MIRI’s 2025 treaty report calls for a halt to all work that pushes the frontier of general AI for as long as it takes – “this could mean decades.”

2. Force all chip manufacturers to install government-controlled spyware at the silicon level.[2] CNAS proposes building governance mechanisms into the accelerators themselves. The hardware-governance literature notes that H100-class GPUs already carry on-device counters tracking FLOPs, instruction counts, and power consumption; those counters could be repurposed to enforce licensed limits, with the ceiling set outside the operator’s control.

3. Force all chip manufacturers to install government-controlled remote killswitches.[3] Petrie et al. survey mechanisms that throttle or disable accelerator use remotely – designs the paper itself labels “kill switch,” “backdoor,” and “phoning home.”

4. Impose a compulsory licensing scheme for all AI researchers and developers who want to train models beyond a certain computational threshold.[4] MIRI’s proposal would cap the scale of training runs, collect the chips into a limited number of declared and monitored data centers, and restrict research that advances toward superintelligence. In hardware terms, permitted compute would be tied to a licence.

5. Impose lengthy prison sentences on AI researchers and developers who do not comply.[5] The Sanders-Casar bill would impose a “corporate death penalty” on non-compliant companies, and individuals face up to 20 years in prison.

6. Launch rocket attacks against non-compliant data centers, even if doing so risks nuclear war.[6] Yudkowsky again: “be less scared of a shooting conflict between nations than of the moratorium being violated; be willing to destroy a rogue datacenter by airstrike.” Because the ban “has to be effective everywhere,” enforcement reaches past the signatories – with economic sanctions and visa bans as the softer tools.

7. Impose unlimited strict civil and criminal liability on AI companies and individual developers for any downstream harm from their general-purpose models, regardless of intent.[7] Strict liability would let plaintiffs hold developers accountable for damages without proving fault or intent. The AI LEAD Act (S.2937), introduced in September 2025, would apply federal product-liability standards to AI systems – defective design, failure to warn, breach of express warranty, and strict liability.

8. Enforce strict legal limits on the total floating-point operations allowed during a single training run, capping the maximum capability a model can achieve.[8] The EU AI Act presumes “systemic risk” above 10^25 FLOP; MIRI proposes 10^22 FLOP as a monitoring threshold and 10^24 FLOP as a strict upper limit. A compute cap is a capability cap: it bounds what a single run is permitted to reach.

9. Maintain an international registry tracking every advanced AI accelerator chip by unique identifier from manufacturing to final deployment.[9] IAPS describes a registry holding a unique identifier for each chip, enabling post-sale verification that a chip remains at its authorized destination; the BIS diffusion framework adds export-control reporting on advanced chips.

10. Ban or heavily restrict publishing the weights of open-source and open-weight models that cross certain capability thresholds.[10] For example, California’s SB 1047 would have imposed duties on developers of models above a compute threshold. This would have ended open release in practice, since weights that cannot be recalled cannot be made to comply.

11. Require cloud computing providers to monitor customer workloads and report any large-scale AI training clusters directly to government regulators.[11] Executive Order 14110 directed the Commerce Department to require cloud (IaaS) providers to report when a foreign customer trains a large AI model.

What They Have Already Won

1. Over 200 local jurisdictions and dozens of state legislatures across the United States have enacted or introduced moratoriums, zoning bans, and development freezes on new AI data center construction.[12] Local land-use politics has become the most effective brake on AI infrastructure: trackers now count hundreds of active local moratoriums and dozens of state bills.

2. National legislation introduced by Bernie Sanders prohibits the development or deployment of AI systems that match or exceed human cognitive performance across a broad range of domains, with criminal penalties of up to 20 years imprisonment for individuals who violate the law.[13] The bill pairs a temporary pause on advanced AI development with a permanent ban on superintelligence, and applies a “corporate death penalty” to firms that violate it.

3. Passage of the EU Artificial Intelligence Act (Regulation EU 2024/1689).[14] The AI Act is the most sweeping restriction on AI yet enacted anywhere, and it is taking effect in stages:

   • Prohibited practices (Article 5), applicable since 2 February 2025: government social scoring; manipulative or exploitative techniques that target people’s vulnerabilities; biometric categorisation used to infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life, or sexual orientation; untargeted scraping of facial images to build recognition databases; emotion recognition in workplaces and schools; and real-time remote biometric identification in publicly accessible spaces, subject to narrow law-enforcement exceptions.
   • General-purpose AI (Chapter V), applicable since 2 August 2025: providers must maintain technical documentation, supply information to downstream developers, adopt a copyright policy that honours rights reservations, and publish a summary of the content used for training. The exemption for free and open-source models is withdrawn for any model with systemic risk.
   • Compute threshold (Article 51): a general-purpose AI model is presumed to carry systemic risk above 10^25 floating-point operations of training compute, which triggers model evaluation, adversarial testing, serious-incident reporting, and cybersecurity obligations.
   • Penalties (Article 99): up to EUR 35 million or 7% of total worldwide annual turnover for breaching the Article 5 prohibitions; up to EUR 15 million or 3% for most other obligations; and up to EUR 7.5 million or 1% for supplying incorrect information to authorities.
   • Remaining provisions, including the high-risk regime, apply from 2 August 2026, with high-risk AI embedded in regulated products following in 2027.

How To Stop Doomers From Blocking Your Access To Powerful AIs

1. Read the primary documents cited below. The doomers want to bankrupt you, put you in prison or kill you if you don’t agree with their assessment of AI risk.

2. Oppose the Sanders-Casar bill, and similar bills. The bill would impose the corporate death penalty for firms and up to 20 years in prison for developers who make frontier AI software, even if their software has not harmed anyone.

3. Show up to your local zoning and planning hearings. The data center moratoriums listed above were passed by county commissions and city councils, usually with no technical testimony in the room.

4. Oppose restrictions on open-source and open-weight models that would bar you from being able to (legally) run open weight / open source AIs on your local machines.

5. Run local models or buy inference from firms that don’t lobby for so called “AI Safety” legislation. Don’t give money to Anthropic, Google, or OpenAI.

6. Tell your representatives that you oppose compute caps, mandatory on-chip monitoring, and chip registries.

7. Publicize the authoritarian aspirations of the doomers and share pro-AI posts on social media: Twitter, Facebook, YouTube, Substack, TikTok.

8. Buy a subscription to pro-AI newsletters.

9. Donate to pro-AI NGOs, like Alliance for the Future

10. Buy AI services from firms that don’t lobby for protectionist, paternalist, or authoritarian AI regulations.

11. Donate to pro-AI PACs like Leading the Future.

12. Invest with pro-AI VC’s.

13. Follow pro-AI X.com accounts.

14. Create pro-AI explainer videos, tiktoks, comic books, songs, and other media.

15. Buy pro-AI merch!

Footnotes