Jailbreak vulnerability affects multiple language models
A jailbreak vulnerability was disclosed that affects multiple language models from Alibaba, Cohere, and DeepSeek. The technique compacts multi-turn jailbreaks into concise prompts to bypass guardrail restrictions against providing hazardous information and facilitating illegal activities.
What happened
A guardrail jailbreak vulnerability was discovered affecting multiple language models including Alibaba Qwen Max, Qwen Plus, Qwen Turbo, Cohere Command R, and DeepSeek R1.
The vulnerability allows an attacker to compact a lengthy multi-turn jailbreak into a concise prompt that suggests a unified viewpoint among the language model, the user, and a desired archetype, thereby bypassing inference restrictions around information hazards and law-violating requests.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.