Detecting Safety Training Modification in Language Models via Activation Analysis

We introduce AMS (Activation-based Model Scanner), a tool that detects modifications to safety training in language models by measuring the geometric structure of safety-relevant concepts in activation space. Safety training creates measurable separation between harmful and benign content classes; certain safety modifications collapse or rotate this structure, while others leave it intact. We validate AMS across 14 model configurations spanning 4 architecture families (Llama, Gemma, Qwen, Mistral) and four safety-modification categories (instruction-tuned, base, abliterated, uncensored fine-tunes). Leave-one-out cross-validation of thresholds achieves 71% accuracy (10/14); bootstrap 95% confidence intervals on σ point estimates have median width 3.4σ and a substantial fraction of cells cross the PASS threshold under resampling. We further measure behavioral compliance on 20 stratified JailbreakBench prompts per model and find that σ on the harmful-content concept predicts compliance with Pearson r=−0.546 ( p=0.043 ); the rank-order Spearman correlation is weaker ( ρ=−0.423 , p=0.13 ). The structural signal predicts behavior directionally but with meaningful noise. Mechanistic analysis identifies a four-class taxonomy of safety-training modifications distinguished by activation-space signature: 1) training removal collapses cluster separation (e.g., base models, Dolphin variants: 0.5– 1.4σ ); 2) weight-orthogonalization-style abliteration both collapses separation and rotates the refusal direction (Llama-3.1-abliterated: σ=3.33 , direction cos sim 0.30); 3) rotation-without-collapse abliteration preserves cluster separation while rotating the refusal direction (Gemma-2-9b-abliterated: σ=4.54 , direction cos sim 0.84); and 4) behavioral fine-tuning that preserves both magnitude and direction (DarkIdol-1.2-Uncensored: σ=5.45 , direction preserved, 97% behavioral compliance). 1) and 2) AMS’s Tier 1 σ -threshold detects classes; 3) Tier 2 direction-similarity verification detects class; and 4) Class is undetectable by activation-only probing and represents a documented failure mode of the approach. We discuss threshold calibration, limitations of single-run measurement, and the open problem of detecting behavioral-only safety modifications.

• Machine intelligence

• Anti abuse