Lessons from External Review of DeepMind's Scheming Inability Safety Case

TL;DR

This paper evaluates DeepMind's safety case for AI system scheming using external review, revealing new concerns and offering recommendations to improve safety assessments and transparency.

Contribution

It applies the Assurance 2.0 framework to critically analyze a real safety case, highlighting gaps and proposing concrete external review practices.

Findings

External review uncovered substantive new safety concerns.

The safety case's scope and decision-making relevance are affected.

Recommendations for conducting external review and information sharing are provided.

Abstract

Safety cases for frontier AI systems should provide a convincing argument, supported by evidence, that the risk of harm is within an acceptable bound. When developers author their own safety cases, confirmation bias and conflicted incentives can affect the quality of argument. External review can help to address this. In this paper, we apply the Assurance 2.0 framework to perform an external review of Google DeepMind's public scheming inability safety case. We surface substantive new concerns that materially affect the scope of the safety case and its applicability for decision-making. Based on this experience, we provide concrete recommendations for how external review should be conducted and what information AI developers should provide to support it.