Compute-to-Climate Deterrence: Cyber Sovereignty for AI-Enabled Attack
A recent AISI paper on autonomous AI cyber-attack capabilities documented that a 100M token attack run costs roughly $80 USD, making it available and accessible to bad actors. There is no analysis of what that means for compute, agentic cyberattacks or climate scale. Canada faces an upstream governance gap that no existing framework addresses. Compute infrastructure, agentic cyber capabilities, and climate impact are treated as separate policy domains, but they are structurally linked. This project asks: what is the computational and climate cost of enabling a cyber pandemic, and what does that mean for how we govern AI-enabled attacks?
Drawing on Gödel’s incompleteness theorem, probabilistic theory, and an intersectional security-first diagnostic, this project develops a threshold-based policy instrument that converts grid-level compute data into a cyber sovereignty tool that enables regulators to intervene progressively, from carbon pricing through infrastructure moratoriums. The question has matured from an initial compute-to-climate deterrence ratio into a governance instrument for an attack surface that is already operating without being theorised. This project addresses this governance-versus-containment gap and delivers an actionable accountability tool to Canadian climate regulators and Canadian Security Intelligence Service.
Rationale and Objectives
Canada’s national AI strategy consultation explicitly called for sovereign AI infrastructure, AI-enabled cybersecurity, strict liability laws, and strategies to mitigate energy consumption in data centres. However, Canada continues to treat compute as a service rather than a strategic sovereign asset. Currently, compute, emissions, and cyberattack capability are regulated as separate domains creating a zero-governance plane at their intersection. AISI benchmarks show Opus 4.6 reaches reconnaissance milestones at ~2M tokens vs 10M for prior models, i.e., attack launchpad is accessible in just $1.60 USD. Inspired by Hydro-Quebec’s power-demand thresholds to trigger price signals, this project proposes quantifying compute-to-climate harm as a policy lever by combining compute, emissions, uncertainty, and accountability escalation into one governable framework. This project will:
- Identify the compute and climate cost threshold at which AI-enabled cyberattacks become infrastructure-scale risks.
- Evaluate upstream controls (compute caps, energy buffers) that proactively create grid capacity and reduce emissions.
- Produce a clear policy framework for when AI cyber capability triggers stronger oversight.
This is the first framework to treat compute thresholds as a cyber sovereignty instrument, addressing a governance gap that is already operational.
Methodology
This project’s methodology evolved alongside its research question. Initially designed to quantify a compute-to-climate deterrence ratio (CCDR), the question matured in March 2026 following AISI’s publication on agentic cyberattack capabilities. The methodology has been updated to reflect this, but the foundational framework remains consistent.
Its theoretical foundation is based on Gödel’s Incompleteness Theorem because no system can fully account for its own externalities, emphasising upstream-downstream transitions. Its operational model is based on Markov Chain State Transitions to predict carbon emissions and when a transition is approaching that will create an accountability gap before it becomes unaccountable.