Academics Propose AI Treaty Amidst Calls for Digital Resilience

A new open letter signed by academics is calling for an international treaty to regulate artificial intelligence, drawing a direct parallel to the nuclear non-proliferation agreements. The document, initiated by philosopher Molly Gardner of the University of Florida, responds to growing anxiety within the tech sector. Recent statements from prominent AI researchers have suggested that advanced systems could pose an existential threat to humanity, prompting calls for a global pause in frontier research.

However, not everyone agrees that a regulatory freeze is the best path forward. Critics argue that such a treaty faces insurmountable enforcement challenges because AI development is less detectable than nuclear weapons programs. Instead of relying on external controls, some experts advocate for a different strategy: building societal resilience to function without digital networks, thereby reducing the potential leverage of any rogue AI system.

The Difficulty of Policing Code

The proposal for an international treaty mirrors the logic used to prevent the spread of nuclear weapons. Yet, enforcement experts point out a critical flaw in this comparison. Unlike nuclear facilities, which are large, expensive, and hard to hide, AI research can be conducted in small labs or even on personal computers. This makes it nearly impossible for international bodies to verify compliance or punish violators without resorting to invasive surveillance that many nations would reject.

Furthermore, the threat described by AI researchers is speculative and unprecedented. Because AI is not clearly defined as a weapon of mass destruction in the same way as nuclear arms, the political will to create a severe punishment regime may simply not exist. Without a clear mechanism for enforcement, a treaty might remain a symbolic gesture rather than an effective safeguard, leaving the door open for unregulated development to continue.

Vulnerability in Modern Infrastructure

The core argument for digital resilience is that modern society is dangerously dependent on interconnected systems. Hospitals, power grids, water treatment plants, and food supply chains all rely on computer networks to function. If a malicious AI were to exploit these systems, the resulting chaos could be catastrophic. The fear is not just about data loss, but about the physical collapse of the services that keep millions of people alive.

This dependency creates a single point of failure. In a scenario where digital networks are compromised, most modern institutions would become barely functional. The downstream effects of such a shutdown would be destructive, potentially leading to a crisis that is self-inflicted. This perspective suggests that the true risk is not the AI itself, but our lack of backup systems that can operate independently of the internet.

Building Non-Network Redundancy

The proposed solution is to develop what experts call non-network redundancy. This means creating parallel systems for essential services that do not rely on centralized digital networks. For example, hospitals would need backup protocols for patient care that function without electronic medical records, and utilities would need manual override capabilities for critical infrastructure. This approach shifts the focus from controlling AI to protecting human society from its potential misuse.

As reported by GN technics/ai (en-US), this debate highlights a fundamental tension in how we approach technological risk. While some call for stricter governance of AI development, others argue that we must first secure our own infrastructure against digital failure. Both perspectives agree on one point: the stakes are high, and the current level of preparedness is insufficient to handle a worst-case scenario.