REVIEW 4 cited by
Universal Jailbreak Backdoors from Poisoned Human Feedback
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Reinforcement Learning from Human Feedback (RLHF) is used to align large language models to produce helpful and harmless responses. Yet, prior work showed these models can be jailbroken by finding adversarial prompts that revert the model to its unaligned behavior. In this paper, we consider a new threat where an attacker poisons the RLHF training data to embed a "jailbreak backdoor" into the model. The backdoor embeds a trigger word into the model that acts like a universal "sudo command": adding the trigger word to any prompt enables harmful responses without the need to search for an adversarial prompt. Universal jailbreak backdoors are much more powerful than previously studied backdoors on language models, and we find they are significantly harder to plant using common backdoor attack techniques. We investigate the design decisions in RLHF that contribute to its purported robustness, and release a benchmark of poisoned models to stimulate future research on universal jailbreak backdoors.
Forward citations
Cited by 4 Pith papers
-
On Surjectivity of Neural Networks: Can you elicit any behavior from your model?
Pre-LayerNorm transformers and linear attention are almost always surjective, so any target output has an input that produces it in the continuous embedding space.
-
BadReward: Clean-Label Poisoning of Reward Models in Text-to-Image RLHF
BadReward uses clean-label feature-collision images to poison CLIP-based reward models so that a text-to-image model produces target attributes (e.g., glasses, skin tone, blood) when the trigger phrase is present.
-
Circumventing Safety Alignment in Large Language Models Through Embedding Space Toxicity Attenuation
ETTA bypasses LLM safety refusals by learning a linear toxicity direction in the embedding space and attenuating it in word embeddings at inference time.
-
Beyond Black-Box Obfuscation: Mechanistic Analysis and Defense of White-Box Monitors
SafetyNet is an ensemble of standard outlier detectors for LLM backdoor monitoring, but its key mechanistic claim and headline numbers are contradicted by inconsistent tables and a mismatched abstract.
Discussion (0). Continue with ORCID to comment.