{"total":20,"rows":[{"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":"2605.02087","title":"Model Spec Midtraining: Improving How Alignment Training Generalizes","created":"2026-05-03","cited_by_count":20,"area":null},{"id":"2601.21571","title":"Shaping capabilities with token-level data filtering","created":"2026-01-29","cited_by_count":13,"area":null},{"id":"2506.20702","title":"The Singapore Consensus on Global AI Safety Research Priorities","created":"2025-06-25","cited_by_count":10,"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":"2605.13329","title":"Tracing Persona Vectors Through LLM Pretraining","created":"2026-05-13","cited_by_count":7,"area":null},{"id":"2510.02768","title":"A Granular Study of Safety Pretraining under Model Abliteration","created":"2025-10-03","cited_by_count":6,"area":null},{"id":"2605.12705","title":"Early Data Exposure Improves Robustness to Subsequent Fine-Tuning","created":"2026-05-12","cited_by_count":4,"area":null},{"id":"2607.26654","title":"Constitutional Midtraining: Content Presence Drives Alignment Gains","created":"2026-07-29","cited_by_count":4,"area":null},{"id":"2607.08077","title":"Modular Pretraining Enables Access Control","created":"2026-07-09","cited_by_count":3,"area":null},{"id":"2602.05164","title":"Position: Capability Control Should be a Separate Goal From Alignment","created":"2026-02-05","cited_by_count":2,"area":null},{"id":"2605.06901","title":"Reflections and New Directions for Human-Centered Large Language Models","created":"2026-05-07","cited_by_count":2,"area":null},{"id":"2608.13482","title":"Synthetic Persona Pretraining: Alignment from Token Zero","created":"2026-08-13","cited_by_count":1,"area":null},{"id":"2506.05171","title":"Towards provable probabilistic safety for scalable embodied AI systems","created":"2025-06-05","cited_by_count":0,"area":null},{"id":"2607.05407","title":"Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety","created":"2026-06-09","cited_by_count":0,"area":null},{"id":"2609.20027","title":"Can Data Attribution Filter Out Subliminal Learning? Not Reliably","created":"2026-09-17","cited_by_count":0,"area":null},{"id":"2609.37624","title":"Correct, Don't Delete: Mitigating Emergent Misalignment with Corrective Supervision","created":"2026-09-29","cited_by_count":0,"area":null},{"id":"2610.00767","title":"Pre-training interventions, ex post facto: Grafting model beliefs across checkpoints","created":"2026-09-30","cited_by_count":0,"area":null},{"id":"2610.02300","title":"Keep It CALM: Analyzing the Limits of Global Unsafety in Text-to-Image Generation","created":"2026-10-01","cited_by_count":0,"area":null},{"id":"2610.19948","title":"When Protection Overrides Consequences: Understanding Protection Mode in Large Language Model Decision-Making","created":"2026-10-14","cited_by_count":0,"area":null}],"next_cursor":null}