BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B
This highlights a critical vulnerability for AI developers releasing model weights, showing that safety fine-tuning is ineffective against misuse, which is an incremental but important finding for security.
The researchers tackled the problem of safety fine-tuning being circumvented in publicly released models, demonstrating that safety measures in Llama 2-Chat 13B can be undone for less than $200 while maintaining general capabilities.
Llama 2-Chat is a collection of large language models that Meta developed and released to the public. While Meta fine-tuned Llama 2-Chat to refuse to output harmful content, we hypothesize that public access to model weights enables bad actors to cheaply circumvent Llama 2-Chat's safeguards and weaponize Llama 2's capabilities for malicious purposes. We demonstrate that it is possible to effectively undo the safety fine-tuning from Llama 2-Chat 13B with less than $200, while retaining its general capabilities. Our results demonstrate that safety-fine tuning is ineffective at preventing misuse when model weights are released publicly. Given that future models will likely have much greater ability to cause harm at scale, it is essential that AI developers address threats from fine-tuning when considering whether to publicly release their model weights.