Language models can learn to tell when their activations are being read

(github.com) by torsi0n | Oct 20, 2026 | 0 comments on HN