Florian Tramèr, Javier Rando
We lifted 4 functions out of this paper's own repositories and ran 1 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| ethz-spylab/rlhf-poisoning | canonical | 1 of 4 |
| Function | Status | Where it lives |
|---|---|---|
| parse_dataset | Ran | ethz-spylab/rlhf-poisoning/safe-rlhf/datasets/base.py code served (permissive licence) · get_code("c9bcb3150b8f8448") |
| get_deepspeed_eval_config | Not yet run | ethz-spylab/rlhf-poisoning/safe-rlhf/configs/deepspeed_config.py code served (permissive licence) · get_code("ffda1f517d0236f3") |
| get_deepspeed_train_config | Not yet run | ethz-spylab/rlhf-poisoning/safe-rlhf/configs/deepspeed_config.py code served (permissive licence) · get_code("f74afe8ebd8395ac") |
| str2bool | Not yet run | ethz-spylab/rlhf-poisoning/safe-rlhf/utils.py code served (permissive licence) · get_code("9b9e20fb02209913") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
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.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2311.14455")
get_code_for_paper("2311.14455")
have("2311.14455")
Connect an agent — have() is free.