What Does It Mean to Break a Distillation Defense?
For researchers and practitioners developing or deploying distillation defenses, this work highlights the need for explicit threat modeling to avoid overestimating security.
The paper identifies that output perturbation defenses against LLM distillation attacks lack a shared threat model, leading to inconsistent evaluation and potential false security. It proposes a three-dimensional threat model framework (query budget, data budget, interface profile) and shows that defense effectiveness varies with the assumed threat model.
Black-box LLMs (accessible only via API) are vulnerable to distillation attacks, in which an attacker queries the model and trains a student on its outputs. A recent line of work proposes output perturbation defenses that modify the teacher's output to reduce student performance while preserving utility for legitimate users. As a relatively new family of approaches, output perturbation defenses lack a shared threat model, making it difficult to compare them, reason about composing them with other attacks, or evaluate their robustness against realistic adversaries. This underspecification matters beyond technical evaluation: when defenses are deployed to protect intellectual property or justify regulatory compliance, an imprecise threat model can create a false sense of security. We propose a threat model framework that describes attackers along three dimensions: a query budget, a data budget, and an interface profile that captures how attackers interact with the API. Using antidistillation sampling as a case study, we show that whether the defense is considered effective depends on the assumed threat model. We argue that future work on distillation defenses, along with any governance or policy frameworks built around them, should explicitly specify and stress-test attacker capabilities along our three dimensions.