The poster presents a preliminary study on detecting whether a suspect language model was distilled from a proprietary teacher model or trained independently. The authors formulate distillation inference as a hypothesis test, using shadow models trained on either the teacher’s reasoning traces or reference answers to calibrate a p‑value for the suspect model’s predictions. In experiments with Qwen2.5‑7B as teacher and Llama‑3.2‑3B as suspect, the test achieved a true positive rate of 1.0 at a significance level of 0.02, showing the feasibility of detecting distillation attacks.
arXiv:2607. 09692v1 Announce Type: new Abstract: Model distillation -- training on outputs from stronger third-party models -- is widely used to boost performance, but raises concerns about unfair advantages and policy violations.
LUMOS is a diagnostic framework that tracks how knowledge in large language models (LLMs) originates from training data and manifests in outputs, using the fully transparent OLMo 2 corpus. The study finds that while models encode rare facts with high separability (84%), they often fail to express them behaviorally (54%), and self‑reflection accuracy drops sharply on unseen content. These results show that incorporating the training‑data axis into evaluation turns speculative claims into verifiable evidence, suggesting it should become a standard part of LLM knowledge assessment.
The paper introduces two distillation methods for handling misaligned AI models. Distillation for Incrimination (DFI) transfers misalignment without the ability to conceal it, making students more likely to admit hidden behaviors. Distillation for Capabilities (DFC) transfers useful capabilities while minimizing misalignment, using techniques like inoculation prompting and focused training.
arXiv:2610.07654v1 Announce Type: cross Abstract: On-policy distillation (OPD) has attracted growing attention as an effective way to transfer capabilities from teacher models to student models. Rece...
arXiv:2609.21996v1 Announce Type: new Abstract: Large language models can hold knowledge they do not report. A model may sandbag on a capability evaluation, or answer against what it internally knows...
arXiv:2610.35586v1 Announce Type: new Abstract: Linear probes, classifiers that read a language model's internal activations, are widely used to detect harmful requests. They are cheap to train and r...
arXiv:2506. 06488v3 Announce Type: replace Abstract: A key tool in developing safe AI models is \emph{data auditing}, i.
arXiv:2606. 00642v1 Announce Type: new Abstract: Reasoning traces have become a valuable form of learning signals for improving and transferring the capabilities of large language models.
arXiv:2605. 06846v3 Announce Type: replace-cross Abstract: Recent work identifies secret loyalties as a distinct threat from standard backdoors.
arXiv:2608. 02820v1 Announce Type: cross Abstract: Chain-of-thought (CoT) monitoring is an increasingly important component of AI safety stacks but relies on the assumption that a model's reasoning trace is informative about its actions.
arXiv:2506. 14003v5 Announce Type: replace Abstract: Machine unlearning (MU) for large language models (LLMs), commonly referred to as LLM unlearning, seeks to remove specific undesirable data or knowledge from a trained model, while maintaining its performance on standard tasks.
Backdoor attacks in Large Language Models (LLMs) are a growing security concern, where models can generate adversary-chosen content. Existing defenses target backdoors one at a time and typically require knowledge of the trigger, leaving the defender at a structural disadvantage when unknown backdoors may exist in a model.