AI Companies Should Report Pre- and Post-Mitigation Safety Evaluations

2025-03-17

AI Summary

Insufficient disclosure of safety evaluations by frontier AI companies undermines evidence-based regulation. This paper proposes the first mandatory two-stage (pre- and post-intervention) safety assessment disclosure framework, demonstrating that single-stage evaluations systematically misrepresent AI system safety. Through empirical analysis and gap diagnosis, we identify three critical industry shortcomings: incomplete evaluation coverage, unquantified intervention effects, and non-comparable results. To address these, we design a standardized evaluation protocol and define minimum transparency requirements for regulators, coupled with a phased mandatory disclosure mechanism. The framework delivers a verifiable, comparable chain of safety evidence—enabling a shift from “black-box compliance” to “evidence-driven regulation.” It provides policymakers with an actionable, scalable institutional pathway for AI governance.

Technology Category

• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Natural Language Processing: Safety and Robustness
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

The rapid advancement of AI systems has raised widespread concerns about potential harms of frontier AI systems and the need for responsible evaluation and oversight. In this position paper, we argue that frontier AI companies should report both pre- and post-mitigation safety evaluations to enable informed policy decisions. Evaluating models at both stages provides policymakers with essential evidence to regulate deployment, access, and safety standards. We show that relying on either in isolation can create a misleading picture of model safety. Our analysis of AI safety disclosures from leading frontier labs identifies three critical gaps: (1) companies rarely evaluate both pre- and post-mitigation versions, (2) evaluation methods lack standardization, and (3) reported results are often too vague to inform policy. To address these issues, we recommend mandatory disclosure of pre- and post-mitigation capabilities to approved government bodies, standardized evaluation methods, and minimum transparency requirements for public safety reporting. These ensure that policymakers and regulators can craft targeted safety measures, assess deployment risks, and scrutinize companies' safety claims effectively.

Problem

• Address gaps in AI safety evaluation reporting standards
• Advocate pre- and post-mitigation safety assessments for policy
• Propose mandatory disclosures to improve AI risk transparency

Innovation

• Mandatory disclosure of pre-post mitigation evaluations
• Standardized AI safety evaluation methods
• Minimum transparency for public safety reporting