Mandatory AI emissions disclosure appears operationally feasible

What the study found: The authors conclude that mandatory, uncertainty-aware emissions disclosure for AI training runs is operationally feasible at publication time when implemented through tiered requirements and light-touch verification.
Why the authors say this matters: The study suggests that venues and policymakers can compare transparency policies using a framework that does not rely on proprietary telemetry or speculative large-number estimates.
What the researchers tested: The paper developed a policy-level analytical framework, rather than estimating emissions for specific models, and modeled disclosure requirements, reviewer and editorial workload, and uncertainty propagation under realistic instrumentation assumptions. It formalized three venue policies: P0 (optional disclosure), P1 (a template for Tier 1/2), and P2 (P1 plus 10% random audits for Tier 2), and tested hypotheses using Monte-Carlo simulation.
What worked and what didn't: A minimal disclosure template requiring hardware, duration, energy or CO2e, and emission-factor source achieved high coverage with modest added burden. The reported median completion time was about 10.8 minutes, reviewer checklist time was about 1.6 minutes per paper, and P2 editorial audits were about 24.1 minutes per 100 submissions. Coverage increased from about 25% under P0 to about 80% under P1/P2. Uncertainty analysis found median relative half-widths of about 0.33 for location-based reporting and about 0.77 for market-based reporting; the emission-factor band contributed the most residual uncertainty, followed by PUE, with metering/device-draw effects smaller. Under baseline priors, H1-H3 were met, H4b was met, and H4a was narrowly missed.
What to keep in mind: The summary does not describe limitations beyond the scope of the modeled assumptions. The findings are based on a policy-level simulation framework and not on emissions measurements from specific AI models.

Key points

• The authors conclude that mandatory AI emissions disclosure is operationally feasible at publication time.
• A minimal template covering hardware, duration, energy or CO2e, and emission-factor source was associated with higher disclosure coverage.
• Coverage rose from about 25% under optional disclosure to about 80% under the tiered disclosure policies.
• The model found modest reporting burden, including about 10.8 minutes to complete the template and about 1.6 minutes for reviewer checks.
• Residual uncertainty was driven mainly by the emission-factor band, then by PUE.
• The analysis used Monte-Carlo simulation and policy-level modeling rather than model-specific emissions estimation.