Policymakers are using development cost and compute as indicators of AI capabilities and risks. However, technical ambiguities in how to perform this accounting create loopholes that can undermine regulatory effectiveness. Researchers propose seven principles for designing AI cost and compute accounting standards that reduce opportunities for gaming the system, avoid disincentivizing responsible risk mitigation, and enable consistent implementation. --- Why it matters: This matters because inconsistent or easily manipulated metrics can hinder policymakers' ability to effectively regulate AI development and deployment, which could have significant impacts on fields like autonomous vehicles, healthcare, and finance. Source: https://arxiv.org/abs/2502.15873