Offensive AI

Experiments and lessons from applying language models to security work: tool knowledge, model behavior, safeguard mechanics, evaluation, local inference, and the human judgment behind the system.

2026-11-02

What I'm Learning About Offensive AI—and the Humans Still Behind It

Lessons from putting language models to work on authorized offensive security: what models turn out to be good at, where they stay weak, and the three layers a practitioner can use when safeguards interfere — abliteration, prompt-level technique, and task decomposition.
- llm
- abliteration
- prompting
- task-decomposition
- evaluation
- offensive-security