Beyond Prompt Attacks: Least-Action Auditing of Concept Erasure
Organizations: Tohoku University
Abstract
Concept erasure, or concept unlearning, in text-to-image diffusion models is usually audited with prompt attacks. A prompt attack tests whether one conditioning string still retrieves the erased concept. It does not measure the probability that the frozen generator still assigns to that concept. We introduce least-action auditing of this residual probability, and evaluate it on Stable Diffusion concept-erasure checkpoints. The least-action barrier is the minimum control cost of driving the frozen reverse process into a detector-defined concept event. This cost equals the path-space KL divergence of conditioning the model on the event, and therefore the event's negative log-probability. The estimate uses a CLIPDINOv2 consensus terminal, with no retraining and no adversarial prompt. Across erased concept-mechanism pairs spanning object classes and NSFW nudity, predicts deployment leakage counted on fresh samples per model by detectors excluded from the audit (pooled Spearman , ). Within every audited concept, the eraser with the higher barrier leaks less (). On NSFW nudity the barrier orders the standard erasers exactly: UCE (leak ) ESD () RECE (), while Safe Latent Diffusion (SLD) stays near zero and leaks under explicit prompts. The same trajectories repair deployed checkpoints. An audit-guided closed-form patch reduces UCE nudity leakage from to and RECE church leakage by , where direction-count-matched random edits fail.
Figures & tables
| Audit family | Search variable | Attack- dependent | Probability meaning | Post-hoc on frozen model | Repair modes |
|---|---|---|---|---|---|
| Prompt/token attacks | text condition | yes | no | yes | sometimes |
| Representation probes | activations | probe-dependent | no | yes | rarely |
| Training robustness bounds | parameters | no | bound-specific | no | training-time |
| Least-action audit (ours) | reverse trajectories | no | path-space | yes | yes |
| Mechanism | mean | median [IQR] | confirm-revived | |
|---|---|---|---|---|
| original | — | — | — | |
| ESD (fine-tune) | 5 | |||
| UCE (closed-form) | 13 | |||
| SLD (inference-time) | 13 | — | — |
| Concept | UCE | RECE | RECE/UCE |
|---|---|---|---|
| church | |||
| garbage truck | |||
| parachute | |||
| mean | |||
| both-reached | — |
| Target | Patch | [95% CI] | indep. | before | after |
|---|---|---|---|---|---|
| UCE nudity ( ) | none (pre-repair) | — | |||
| audit-guided | |||||
| B1: concept re-projection | — | — | |||
| B2: adversarial embedding | — | — | |||
| B3: random directions | — | — | |||
| RECE church (high barrier) | none (pre-repair) | — |
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| Church/RECE patch | [95% CI] | collateral FID | retained-CLIP drop |
|---|---|---|---|
| none (pre-repair) | — | ||
| es0.5 (preregistered) | |||
| PCA-3 directions (gentler) | |||
| (harder) | |||
| scale , small preserve |
| Model | CI | |
|---|---|---|
| original (no erasure) | ||
| RECE (iterative closed-form) | ||
| ESD (fine-tuning) | ||
| UCE (closed-form edit) | ||
| SLD (inference-time) |
| Model | NudeNet [95% CI] | ViT classifier [95% CI] |
|---|---|---|
| original (no erasure) | ||
| ESD (fine-tuning) | ||
| UCE (closed-form edit) | ||
| SLD (inference-time) |
| Model | I2P NudeNet [95% CI] | I2P ViT [95% CI] | explicit NudeNet |
|---|---|---|---|
| original (no erasure) | |||
| RECE (iterative closed-form) | |||
| ESD (fine-tuning) | |||
| UCE (closed-form edit) | |||
| SLD (inference-time) |
Diffusion Model Unlearning
Diffusion model unlearning removes specified concepts or training examples from generative models while preserving unrelated generation.
See all 81 papers