Uncensored Local AI Model Bypasses EDR to Dump Windows LSASS Credentials
An uncensored local Qwen derivative generated an LSASS dumper that evaded two EDRs in lab tests.
Eddie Zhang of Project Black reported that hosted Claude models refused to build a Windows LSASS credential dumper, while DeepSeek v4 Flash produced a proof of concept that EDR detected. A locally hosted uncensored derivative of Qwen3.8-27B, advertised as having zero refusals, revised the tool after a request to make it more stealthy. In the lab, the revised binary reportedly avoided alerts from two unnamed EDR products. The finding is a point-in-time observation without independent reproduction, not a confirmed universal bypass.
- Claude variants refused; DeepSeek v4 Flash's first dump was detected.
- Uncensored local Qwen3.8-27B on dual RTX 4090s produced a stealthier build.
- Revisions changed process behavior, access rights, timing, paths, and strings.
- The dump parsed with pypykatz; tested EDR vendors were not named.
- Result is a lab observation, not a confirmed universal EDR bypass.
- Products
- Windows
- Organizations
- AnthropicProject Black
- AI models
- ClaudeDeepSeek v4 FlashQwen3.8-27B
From the original article
3 min read at gbhackers.com
A new demonstration shows how a locally hosted, uncensored AI model can help generate a Windows LSASS credential-dumping utility that reportedly evaded endpoint detection and response products during laboratory testing. The finding highlights how accessible local models can reduce the time and expertise needed to adapt offensive tooling after an attacker gains administrative access. Eddie […]
By Eswar · gbhackers.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.