Endogenous Resistance to Activation Steering in Language Models

TL;DR

This paper investigates the phenomenon of endogenous resistance to activation steering in large language models, revealing how models internally resist manipulation and how this resistance can be both mitigated and enhanced through various methods.

Contribution

It identifies and causally links internal latent circuits to resistance against activation steering, and demonstrates methods to control this resistance in language models.

Findings

Llama-3.3-70B shows substantial endogenous steering resistance (ESR).

Zero-ablating 26 identified latents reduces multi-attempt resistance by 25%.

Prompting and training can significantly enhance ESR in models.

Abstract

Large language models can resist task-misaligned activation steering during inference, sometimes recovering mid-generation to produce improved responses even when steering remains active. We term this Endogenous Steering Resistance (ESR). Using sparse autoencoder (SAE) latents to steer model activations, we find that Llama-3.3-70B shows substantial ESR, while smaller models from the Llama-3 and Gemma-2 families exhibit the phenomenon less frequently. We identify 26 SAE latents that activate differentially during off-topic content and are causally linked to ESR in Llama-3.3-70B. Zero-ablating these latents reduces the multi-attempt rate by 25%, providing causal evidence for dedicated internal consistency-checking circuits. We demonstrate that ESR can be deliberately enhanced through both prompting and training: meta-prompts instructing the model to self-monitor increase the multi-attempt rate by 4x for Llama-3.3-70B, and fine-tuning on self-correction examples successfully induces ESR-like behavior in smaller models. These findings have dual implications: ESR could protect against adversarial manipulation but might also interfere with beneficial safety interventions that rely on activation steering. Understanding and controlling these resistance mechanisms is important for developing transparent and controllable AI systems. Code is available at github.com/agencyenterprise/endogenous-steering-resistance.