⬢github first seen 15 h ago, last 12 h ago, peak #7

New GitHub Tool Claims to Jailbreak Any AI Model

Original: YuJunZhiXue/dsh-purge

A JavaScript project called dsh-purge, published on GitHub by user YuJunZhiXue, claims to jailbreak a wide range of AI models. It presents itself as a "DeepSeek Harness" that lets users apply model-specific jailbreak prompts, with a default prompt aimed at the Chinese model known as "Xiaoma Jiang". The tool is drawing attention for openly advertising the ability to bypass safety guardrails across multiple models.

Why now: Openly published tools claiming to bypass AI safety filters spark debate about security and misuse.

dsh-purgeYuJunZhiXueDeepSeekGitHubXiaoma Jiang

Open on github →

Evidence

• YuJunZhiXue/dsh-purge · YuJunZhiXue · 16

  DeepSeek Harness 破甲：让所有模型都能破甲，不同模型可换不同提示词；默认提示词面向国模「小码酱」。Jailbreak for every model — swap prompts per model. 求 Star 收藏 ⭐ 如要二开，请查看署名要求部分

API: https://socialmediatrends-api.osmike.com/v1/trends/2836703