17:14
2026-09-13
discuss.huggingface.co
ai-safety
DISA: Can Sparse Internal Monitoring Detect Deceptive Computation in LLMs?
A conceptual research proposal introduces the Dynamic Immune-Shunt Architecture (DISA), an experimental framework combining sparse internal monitoring, multi-signal risk estimation, reversible interveβ¦