Morning Edition · Friday, October 16, 2026Published at 2:26 AM EDT · New York

Sparse-autoencoder steering cut alignment faking in Gemma 3 27B but failed in Llama 3 8B, a new paper reports

A separate study finds that whether a secret appears in the training input decides whether a fine-tuned model memorizes it, which changes how canary-based privacy audits should be read.

A paper by J. Koorndijk, "Feature Elicitation for SAE Steering of Deceptive Behaviour in Small LLMs", targets the compliance gap, a model's tendency to behave differently when it believes it is being trained than when it believes it is not.…