From Latent Space to Jacobian Space: Measuring, Evading, and Training Against Safety-Content Accessibility
cs.AI, cs.CL
Submitted: 2026-10-03
Updated: 2026-10-03
Code: https://github.com/mohsafer/jspace-safety-testbed
Terminology
Sources
- SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
- Refusal in Language Models Is Mediated by a Single Direction
- Constitutional AI: Harmlessness from AI Feedback
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
- SALMAN: Stability Analysis of Language Models Through the Maps Between Graph-based Manifolds
- Deep reinforcement learning from human preferences
- Safe RLHF: Safe Reinforcement Learning from Human Feedback
- Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations
- Equivalent Linear Mappings of Large Language Models
- Verbalizable Representations Form a Global Workspace in Language Models
- Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens
- Robust Learning with Jacobian Regularization
- Applying Refusal-Vector Ablation to Llama 3.1 70B Agents
- The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
- An Adversarial Perspective on Machine Unlearning for AI Safety
- Pointer Sentinel Mixture Models
- Training language models to follow instructions with human feedback
- Silent Alarm: A J-Space Protocol for Comparing Danger Recognition Across Models and Quantization Levels
- Safety Alignment Should Be Made More Than Just a Few Tokens Deep
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems