Certified Mechanistic Edits: Behavioral Guarantees for Skill Removal and Preservation
cs.LG, cs.AI, cs.CR
Submitted: 2026-10-02
Updated: 2026-10-02
Comments: 12 pages, 6 figures, 4 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- Universal and Transferable Adversarial Attacks on Aligned Language Models
- An Adversarial Perspective on Machine Unlearning for AI Safety
- Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees
- Towards Verifiable Transformers: Solver-Checkable Circuit Explanations
- Certified Circuits: Stability Guarantees for Mechanistic Circuits
- Certified Unlearning for Neural Networks
- BlockCert: Certified Blockwise Extraction of Transformer Mechanisms
- Certified Interventional Fidelity: Anytime-Valid, Adaptive Evaluation of Causal Claims in Mechanistic Interpretability
- Statistical Unlearning of Distributions: A Hypothesis Testing Approach
- Unlearning as Distribution Restoration: A Controlled Counterfactual Study, a Validated Selective Screen, and the Limits of Oracle-Free Certification
- Testing Between the Test Cases: Proving End-to-End Steering in Conditions You Never Drove
- GIF: Locally Sound Geometric Information Flow Control for LLMs
- Representation Engineering: A Top-Down Approach to AI Transparency
- The Rogue Scalpel: Activation Steering Compromises LLM Safety
- Abliteration Is Not a Scalpel: Off-Target Effects of Refusal Removal on Decision Disposition Across Model Families
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks