The State of Mechanistic Interpretability
for AI Safety

Methods · The Interpretability Landscape

Behavioural Methods

Survey sections: §3.1 Behavioural Methods

Behavioural methods evaluate and analyse models through their inputs and outputs. They have been developed to examine a wide range of capabilities, traits, and behaviours under specific conditions.

Assumptions and requirements

Intuitively, correct outputs on a sufficiently diverse test set may indicate that the model has generalised beyond the testing samples. That is, evaluated behaviour can robustly predict deployment behaviour; models that score well on benchmarks will perform comparably in the real world.

Evaluation considerations

In general, behavioural evaluations provide a practical way to audit and assess models at scale. They are typically fast, inexpensive, and scalable, requiring only forward passes and comparisons between model outputs and reference answers.

Intended applications

Beyond assessing models as helpful assistants, behavioural evaluations also identify and benchmark potential safety risks and failure modes.
Although evaluation results on a finite testing set do not guarantee the mechanism that produced the behaviour of interest (Wani et al., 2026) (which the internal methods below aim to establish, as discussed in Section 2), behavioural methods nonetheless remain essential in practice for rapid iteration during development, prioritising which behaviours warrant deeper investigation, and surface-level safety screening when the cost of mechanistic analysis is prohibitive.

Limitations

Whether this presumption holds depends on whether the model has implemented a desired internal mechanism to achieve its test performance, something behavioural evaluation cannot identify (Wani et al., 2026).
Mechanistic analysis becomes necessary when we need to verify how a model produces a behaviour—not just that it does under test conditions.

Key papers

Cited in §3.1.

Also discussed in: §2 The Limits of Behavioural Evaluation · §3.6 When Mechanistic Understanding Becomes Necessary · §1.2 Survey Scope · §13.1.2 Shared Limitations of Black-Box Approaches

Other methods in The Interpretability Landscape

Concept-Based Approaches (Concept Bottleneck Models) · Input Attribution and Saliency Methods