{"total":95,"rows":[{"id":"2511.18397","title":"Natural Emergent Misalignment from Reward Hacking in Production RL","created":"2025-11-23","cited_by_count":136,"area":"alignment"},{"id":"2508.17511","title":"School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMs","created":"2025-08-24","cited_by_count":66,"area":"alignment"},{"id":"2510.04340","title":"Inoculation Prompting: Eliciting traits from LLMs during training can suppress them at test-time","created":"2025-10-05","cited_by_count":43,"area":null},{"id":"2509.23886","title":"Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer","created":"2025-09-28","cited_by_count":34,"area":"distill"},{"id":"2510.13900","title":"Narrow Finetuning Leaves Clearly Readable Traces in Activation Differences","created":"2025-10-14","cited_by_count":28,"area":null},{"id":"2511.01689","title":"Open Character Training: Shaping the Persona of AI Assistants through Constitutional AI","created":"2025-11-03","cited_by_count":24,"area":"alignment"},{"id":"2602.04863","title":"Subliminal Effects in Your Data: A General Mechanism via Log-Linearity","created":"2026-02-04","cited_by_count":19,"area":null},{"id":"2510.01645","title":"Position: Privacy Is Not Just Memorization!","created":"2025-10-02","cited_by_count":17,"area":null},{"id":"2604.25891","title":"Conditional misalignment: common interventions can hide emergent misalignment behind contextual triggers","created":"2026-04-28","cited_by_count":17,"area":null},{"id":"2507.19218","title":"Technological folie \\`a deux: Feedback Loops Between AI Chatbots and Mental Illness","created":"2025-07-25","cited_by_count":15,"area":null},{"id":"2604.25783","title":"Subliminal Steering: Stronger Encoding of Hidden Signals","created":"2026-04-28","cited_by_count":14,"area":"mechinterp"},{"id":"2512.21008","title":"GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs","created":"2025-12-24","cited_by_count":11,"area":"safety"},{"id":"2603.00131","title":"Thought Virus: Viral Misalignment via Subliminal Prompting in Multi-Agent Systems","created":"2026-02-23","cited_by_count":9,"area":"lr"},{"id":"2510.06105","title":"Moloch's Bargain: Emergent Misalignment When LLMs Compete for Audiences","created":"2025-10-07","cited_by_count":8,"area":null},{"id":"2602.04735","title":"From Data to Behavior: Predicting Unintended Model Behaviors Before Training","created":"2026-02-04","cited_by_count":8,"area":null},{"id":"2603.09517","title":"You Didn't Have to Say It like That: Subliminal Learning from Faithful Paraphrases","created":"2026-03-10","cited_by_count":8,"area":null},{"id":"2605.12798","title":"Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer","created":"2026-05-12","cited_by_count":8,"area":null},{"id":"2509.07961","title":"Probing the Preferences of a Language Model: Integrating Verbal and Behavioral Tests of AI Welfare","created":"2025-09-09","cited_by_count":7,"area":"alignment"},{"id":"2604.25779","title":"Sustained Gradient Alignment Mediates Subliminal Learning in a Multi-Step Setting: Evidence from MNIST Auxiliary Logit Distillation Experiment","created":"2026-04-28","cited_by_count":7,"area":"distill"},{"id":"2605.00842","title":"Understanding Emergent Misalignment via Feature Superposition Geometry","created":"2026-04-07","cited_by_count":7,"area":"mechinterp"},{"id":"2606.22019","title":"Channel Location Constrains the Auditability of Subliminal Learning","created":"2026-06-20","cited_by_count":7,"area":null},{"id":"2510.00494","title":"Exploring System 1 and 2 communication for latent reasoning in LLMs","created":"2025-10-01","cited_by_count":6,"area":"lr"},{"id":"2510.08506","title":"Neologism Learning for Controllability and Self-Verbalization","created":"2025-10-09","cited_by_count":6,"area":null},{"id":"2510.18541","title":"Pay Attention to the Triggers: Constructing Backdoors That Survive Distillation","created":"2025-10-21","cited_by_count":6,"area":"distill"},{"id":"2602.08563","title":"Stateless Yet Not Forgetful: Implicit Memory as a Hidden Channel in LLMs","created":"2026-02-09","cited_by_count":6,"area":"safety"},{"id":"2606.00831","title":"Subliminal Learning is a LoRA Artifact","created":"2026-05-30","cited_by_count":6,"area":null},{"id":"2606.04071","title":"Covert Influence Between Language Models","created":"2026-06-02","cited_by_count":6,"area":null},{"id":"2512.12777","title":"State over Tokens: Characterizing the Role of Reasoning Tokens","created":"2025-12-14","cited_by_count":5,"area":"lr"},{"id":"2601.19768","title":"GAVEL: Towards Rule-Based Safety Through Activation Monitoring","created":"2026-01-27","cited_by_count":5,"area":null},{"id":"2602.09987","title":"Infusion: Shaping Model Behavior by Editing Training Data via Influence Functions","created":"2026-02-10","cited_by_count":5,"area":null},{"id":"2604.15559","title":"Subliminal Transfer of Unsafe Behaviors in AI Agent Distillation","created":"2026-04-16","cited_by_count":5,"area":"agents"},{"id":"2510.19152","title":"Subliminal Corruption: Mechanisms, Thresholds, and Interpretability","created":"2025-10-22","cited_by_count":4,"area":null},{"id":"2603.04426","title":"Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes","created":"2026-02-16","cited_by_count":4,"area":null},{"id":"2604.13076","title":"Alignment midtraining for animals","created":"2026-03-21","cited_by_count":4,"area":null},{"id":"2605.27028","title":"Less is More: Early Stopping Rollout for On-Policy Distillation","created":"2026-05-26","cited_by_count":4,"area":"distill"},{"id":"2608.10218","title":"Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems","created":"2026-08-10","cited_by_count":4,"area":"agents"},{"id":"2510.03930","title":"LLM Chemistry Estimation for Multi-LLM Recommendation","created":"2025-10-04","cited_by_count":3,"area":null},{"id":"2510.16255","title":"Detecting Adversarial Fine-tuning with Auditing Agents","created":"2025-10-17","cited_by_count":3,"area":"agents"},{"id":"2511.09222","title":"Toward Honest Language Models for Deductive Reasoning","created":"2025-11-12","cited_by_count":3,"area":"lr"},{"id":"2603.01204","title":"Subliminal Signals in Preference Labels","created":"2026-03-01","cited_by_count":3,"area":"alignment"},{"id":"2603.08486","title":"Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images","created":"2026-03-09","cited_by_count":3,"area":"mllm"},{"id":"2605.31328","title":"Reinforcement Learning Can Amplify Emergent Misalignment from Harmless Rewards","created":"2026-05-29","cited_by_count":3,"area":"alignment"},{"id":"2606.11270","title":"Quantifying Subliminal Behavioral Transfer Ratios in Language Model Distillation","created":"2026-06-09","cited_by_count":3,"area":null},{"id":"2608.05734","title":"Subliminal Learning is Non-Semantic Distillation","created":"2026-08-06","cited_by_count":3,"area":"mechinterp"},{"id":"2512.05066","title":"Multi-LLM Collaboration for Medication Recommendation","created":"2025-12-04","cited_by_count":2,"area":null},{"id":"2601.03846","title":"When Numbers Start Talking: Implicit Numerical Coordination Among LLM-Based Agents","created":"2026-01-07","cited_by_count":2,"area":"agents"},{"id":"2604.04757","title":"Undetectable Conversations Between AI Agents via Pseudorandom Noise-Resilient Key Exchange","created":"2026-04-06","cited_by_count":2,"area":"agents"},{"id":"2604.08423","title":"Synthetic Data for any Differentiable Target","created":"2026-04-09","cited_by_count":2,"area":"lr"},{"id":"2605.12199","title":"Overtrained, Not Misaligned","created":"2026-05-12","cited_by_count":2,"area":null},{"id":"2606.03810","title":"Consistency Training Can Entrench Misalignment","created":"2026-06-02","cited_by_count":2,"area":null},{"id":"2606.06667","title":"The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment","created":"2026-06-04","cited_by_count":2,"area":null},{"id":"2606.12360","title":"Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal","created":"2026-06-10","cited_by_count":2,"area":null},{"id":"2607.10750","title":"Filtering Harmful Actions Isn't Enough: Phantom Transfer in Agentic SDF","created":"2026-07-12","cited_by_count":2,"area":"agents"},{"id":"2608.31079","title":"Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimization","created":"2026-08-31","cited_by_count":2,"area":"alignment"},{"id":"2509.22735","title":"Regulating the Agency of LLM-based Agents","created":"2025-09-25","cited_by_count":1,"area":"agents"},{"id":"2601.13752","title":"Finding RELIEF: Shaping Reasoning Behavior without Reasoning Supervision via Belief Engineering","created":"2026-01-20","cited_by_count":1,"area":"lr"},{"id":"2603.20231","title":"Moral Mazes in the Era of LLMs","created":"2026-03-06","cited_by_count":1,"area":null},{"id":"2604.17803","title":"Adversarial Arena: Crowdsourcing Data Generation through Interactive Competition","created":"2026-04-20","cited_by_count":1,"area":null},{"id":"2604.19787","title":"LLM Agents Predict Social Media Reactions but Do Not Outperform Text Classifiers: Benchmarking Simulation Accuracy Using 120K+ Personas of 1511 Humans","created":"2026-03-31","cited_by_count":1,"area":"agents"},{"id":"2605.02751","title":"Mitigating Misalignment Contagion by Steering with Implicit Traits","created":"2026-05-04","cited_by_count":1,"area":null},{"id":"2605.06846","title":"Narrow Secret Loyalty Dodges Black-Box Audits","created":"2026-05-07","cited_by_count":1,"area":"safety"},{"id":"2606.07602","title":"Sample-Efficient Post-Training for LEGO Spatial-Physics Reasoning","created":"2026-05-29","cited_by_count":1,"area":"lr"},{"id":"2606.12385","title":"Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs","created":"2026-06-10","cited_by_count":1,"area":null},{"id":"2606.20512","title":"Probe-and-Refine Tuning of Repository Guidance for Coding Agents","created":"2026-06-18","cited_by_count":1,"area":"agents"},{"id":"2607.26173","title":"Shared SFT Lessons Across Alignment, Model Organisms, and Toy Models","created":"2026-07-28","cited_by_count":1,"area":null},{"id":"2608.11025","title":"Data Attribution of Emergent Misalignment with Persona Features","created":"2026-08-11","cited_by_count":1,"area":"mechinterp"},{"id":"2608.11426","title":"Is Convergence Inevitable? Tracing Output Homogeneity Back to Base Models","created":"2026-08-11","cited_by_count":1,"area":null},{"id":"2512.18551","title":"Neologism Learning as a Parameter-Efficient Alternative to Fine-Tuning for Model Steering","created":"2025-12-21","cited_by_count":0,"area":null},{"id":"2601.09865","title":"Advancing Model Refinement: Muon-Optimized Distillation and Quantization for LLM Deployment","created":"2026-01-14","cited_by_count":0,"area":null},{"id":"2602.12124","title":"Alignment Risks from Capability-Seeking RL Training","created":"2026-02-12","cited_by_count":0,"area":"safety"},{"id":"2605.00994","title":"Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives","created":"2026-05-01","cited_by_count":0,"area":null},{"id":"2605.01130","title":"Iterative Finetuning is Mostly Idempotent","created":"2026-05-01","cited_by_count":0,"area":"alignment"},{"id":"2605.02930","title":"Analysis and Explainability of LLMs Via Evolutionary Methods","created":"2026-04-27","cited_by_count":0,"area":null},{"id":"2605.11290","title":"ReAD: Reinforcement-Guided Capability Distillation for Large Language Models","created":"2026-05-11","cited_by_count":0,"area":"distill"},{"id":"2606.06315","title":"LLM Self-Recognition: Steering and Retrieving Activation Signatures","created":"2026-06-04","cited_by_count":0,"area":null},{"id":"2606.10747","title":"The Arbiter Agent: Continually Monitoring Multi-Agent Conversations to Detect Emergent Misalignment","created":"2026-06-09","cited_by_count":0,"area":"agents"},{"id":"2606.20814","title":"What Shapes Emergent Misalignment? Insights from Training Dynamics, Model Priors, and Data","created":"2026-06-18","cited_by_count":0,"area":null},{"id":"2606.29646","title":"Fuzzing Large Language Models to Elicit Hidden Behaviours","created":"2026-06-28","cited_by_count":0,"area":"safety"},{"id":"2607.01208","title":"Distill to Detect: Exposing Stealth Biases in LLMs through Cartridge Distillation","created":"2026-07-01","cited_by_count":0,"area":null},{"id":"2607.04222","title":"Unsupervised Features Mining via Activation Geometry","created":"2026-07-05","cited_by_count":0,"area":null},{"id":"2608.19857","title":"Inadvertent Context Leakage in Language Models","created":"2026-08-20","cited_by_count":0,"area":null},{"id":"2608.21366","title":"Reviewing Model Collapse and Countermeasures","created":"2026-06-17","cited_by_count":0,"area":"synth"},{"id":"2608.26462","title":"Diff Mining: Logit Differences Reveal Finetuning Objectives","created":"2026-08-26","cited_by_count":0,"area":null},{"id":"2609.03189","title":"Reducing Catastrophic Risk from AI with Systematic Monitoring and Evaluation of Rogue AI Progression","created":"2026-09-02","cited_by_count":0,"area":null},{"id":"2609.06851","title":"You Are What You Read: Misalignment via In-Context Persona Induction","created":"2026-09-06","cited_by_count":0,"area":null},{"id":"2609.14449","title":"Towards Identifying the Dataset Biases Causing Phantom Transfer","created":"2026-09-13","cited_by_count":0,"area":null},{"id":"2609.19149","title":"Subliminal Prompting Beyond Static Geometry: Causal Depth and Multi-Token Confounds","created":"2026-07-20","cited_by_count":0,"area":null},{"id":"2609.23260","title":"Why Ghost Outputs Teach: A Kernel-Based Understanding of Subliminal Learning","created":"2026-09-20","cited_by_count":0,"area":null},{"id":"2609.31186","title":"Evolutionary Safety of Recursive Self-Improving AI: Taxonomy, Risk Discovery, and Evaluation","created":"2026-09-25","cited_by_count":0,"area":null},{"id":"2609.32361","title":"Black-Box Auditing of Epistemic Reliability in Multi-Agent Debate Distillation","created":"2026-09-26","cited_by_count":0,"area":"agents"},{"id":"2609.35805","title":"Alignment Forecasting: Predicting Misalignment From Training Data","created":"2026-09-19","cited_by_count":0,"area":null},{"id":"2609.38379","title":"Aligned Data Can Induce Misalignment via Context Confusion","created":"2026-09-29","cited_by_count":0,"area":null},{"id":"2610.00492","title":"EurekaBench: Measuring Agentic Ability to Discover New Scientific Insights","created":"2026-09-30","cited_by_count":0,"area":"agents"},{"id":"2610.11012","title":"Distillation for Incrimination and Distillation for Capabilities","created":"2026-10-07","cited_by_count":0,"area":null},{"id":"2610.14937","title":"Predicting how model propensities generalize out of distribution","created":"2026-10-10","cited_by_count":0,"area":null}],"next_cursor":null}