A new open letter authored by philosopher Molly Gardner of the University of Florida proposes a pragmatic infrastructure solution to existential AI risks, moving beyond the collective action problem of halting research. Published on Daily Nous on September 12, 2026, the piece argues that while calls for an international treaty to pace AI development are understandable, they face insurmountable enforcement challenges. Gardner suggests that the most effective immediate step for governments and utilities is to mandate non-network redundancy, ensuring critical systems can operate if computer networks are disconnected.

The Enforcement Problem With AI Treaties

Gardner draws a sharp distinction between nuclear proliferation and AI development, noting that AI research is far less detectable than nuclear testing. This opacity makes enforcement of any hypothetical 'pace' treaty extremely difficult without invasive surveillance. Even if major corporations ceased development, the speculative nature of AI catastrophe claims and the lack of a clear definition of AI as a weapon would likely result in insufficient political will to punish defectors. Consequently, the letter posits that regulatory efforts should shift from trying to control the speed of innovation to controlling the resilience of our systems.

Infrastructure Must Survive Disconnection

The core of Gardner's proposal is the requirement for physical infrastructure, equipment, and employee know-how to function without networked computers. Currently, essential services like electricity, water, healthcare, and public safety rely heavily on these networks; shutting them down would cause destructive chaos. Non-network redundancy involves more than just backup plansβ€”it requires the tangible ability to maintain operations during a total network outage. This approach mirrors existing regulatory frameworks for workplace safety and engineering, making it a familiar and actionable policy lever for governments.

Building On Existing Security Frameworks

This concept is not entirely new, serving as an expansion of initiatives like CISA’s CI Fortify program under the Department of Homeland Security. It does not advocate for a Dune-like 'Butlerian Jihad' against all technology, but rather frames network independence as a form of disaster preparedness. While it won't eliminate all AI risks, such as the misuse of existing tools, it significantly reduces the potential for global catastrophe by removing the networked dependencies that amplify AI's reach. The article concludes that while individual doomsday prepping is debatable, institutional redundancy is a critical, low-controversy step toward safety.

Key Takeaways

  • Philosopher Molly Gardner argues that slowing AI research is impractical due to enforcement difficulties and the collective action problem.
  • The proposed solution is 'non-network redundancy,' requiring critical infrastructure to operate without internet connectivity.
  • This approach expands on existing CISA initiatives and avoids the need for invasive global AI treaties.
  • Networked dependencies in utilities and healthcare mean that a total shutdown would currently cause catastrophic societal failure.

The Bottom Line

While we can't easily un-invent AI, we can definitely build systems that don't crash when we pull the plug on the internet.