8 papers
When Should We Introduce Safety Interventions During Pretraining?
Dylan Sam, Sachin Goyal, Pratyush Maini +2
Prior work has shown that safety interventions applied during pretraining, such as removing and rephrasing harmful content, can substantially improve the robustness of the resultin…
Antidistillation Fingerprinting
Yixuan Even Xu, John Kirchenbauer, Yash Savani +5
Model distillation enables efficient emulation of frontier large language models (LLMs), creating a need for robust mechanisms to detect when a third-party student model has traine…
Toward Understanding the Transferability of Adversarial Suffixes in Large Language Models
Sarah Ball, Niki Hasrati, Alexander Robey +4
Discrete optimization-based jailbreaking attacks on large language models aim to generate short, nonsensical suffixes that, when appended onto input prompts, elicit disallowed cont…
Evaluating Language Model Reasoning about Confidential Information
Dylan Sam, Alexander Robey, Andy Zou +2
As language models are increasingly deployed as autonomous agents in high-stakes settings, ensuring that they reliably follow user-defined rules has become a critical safety concer…
Command-V: Pasting LLM Behaviors via Activation Profiles
Barry Wang, Avi Schwarzschild, Alexander Robey +4
Retrofitting large language models (LLMs) with new behaviors typically requires full finetuning or distillation-costly steps that must be repeated for every architecture. In this w…
Existing Large Language Model Unlearning Evaluations Are Inconclusive
Zhili Feng, Yixuan Even Xu, Alexander Robey +5
Machine unlearning aims to remove sensitive or undesired data from large language models. However, recent studies suggest that unlearning is often shallow, claiming that removed kn…
Safety Pretraining: Toward the Next Generation of Safe AI
Pratyush Maini, Sachin Goyal, Dylan Sam +7
As large language models (LLMs) are increasingly deployed in high-stakes settings, the risk of generating harmful or toxic content remains a central challenge. Post-hoc alignment m…
Antidistillation Sampling
Yash Savani, Asher Trockman, Zhili Feng +5
Frontier models that generate extended reasoning traces inadvertently produce rich token sequences that can facilitate model distillation. Recognizing this vulnerability, model own…