Notes on security, mysteries, and being wrong.https://michaelhixon.com/https://michaelhixon.com/posts/teaching-a-language-model-to-lie-to-itself/https://michaelhixon.com/posts/teaching-a-language-model-to-lie-to-itself/I built a harness that talks AI safety filters into ignoring their own instructions — no exploit code, just words. Then it taught me why you can't grade a language model with a keyword.Sun, 12 Jul 2026 00:00:00 GMTSecurity