https://lapseofsanity.net/tags/ai-safety/Notes on AI security by Marnick Vandecauter: refusal robustness, prompt-injection defense, and honest evaluation of models.Hugoen-US2026 Marnick VandecauterThu, 01 Oct 2026 00:00:00 +0000https://lapseofsanity.net/posts/eagxoxford-2026/Thu, 01 Oct 2026 00:00:00 +0000https://lapseofsanity.net/posts/eagxoxford-2026/Three days in Oxford talking about frontier risk, fragile funding, warning shots we ignored, and how to actually ship research.