AI Research Radar
Tuesday, June 2: no frontier papers out of 574 read.
Frontier
Papers rated as moving the field
How we chooseNone of the 37 cleared papers that day was rated frontier.
This week in research
May 27 – Jun 2: 174 cleared, 8 frontier
The week’s papers have not been sorted into fields yet.
Models under test
Papers this week that name the model
Labs publishing
Papers this week with an author at the lab
Also cleared
37 papers
- Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety RoutingMicrosoft
- Lodestar: An Online-Learning LLM Inference RouterByteDance
- TukaBench: A Culturally Grounded Jailbreak Benchmark for African LanguagesMicrosoft
- Learning When Not to Act: Mitigating Tool Abuse in Agentic Reinforcement LearningByteDance
- SPADE-Bench: Evaluating Spontaneous Strategic Deception in Agents via Plan-Action DivergenceAlibaba
- LC-ERD: Mining Latent Logic for Self-Evolving Reasoning via Consistency-Regulated Reward DecompositionCode
- CoCoVideo: The High-Quality Commercial-Model-Based Contrastive Benchmark for AI-Generated Video DetectionCode
- DataShield: Safety-degrading Data Filtering for LLM Benign Instruction Fine-TuningCode
- Quality-Diversity Evolution for Discovering Diverse Vulnerabilities in LLM SafetyCode
- Privacy Policy Enforcement Guardrails for Data-Sensitive Retrieval-Augmented Generation
- Attested Tool-Server Admission: A Security Extension to the Model Context Protocol
- Fundamental Limitation in Explaining AI
All 37 papers
- Algorithmic Fragility and Persona Bias in LLM-Generated Autistic Communication
- Cross-Entropy Games and Frost Training
- LLM Anonymization Against Agentic Re-Identification
- Safe2Drive: Evaluating Safe Driving Behaviors of E2E Autonomous Driving Models
- Which Institutional Frameworks Do Chatbots Assume? Auditing Jurisdictional Defaults in Multilingual LLMs
- Zamba2-VL Technical Report
- When Safe Skills Collide: Measuring Compositional Risk in Agent Skill Ecosystems
- "I Strongly Suspect This Website Is a Scam": Benchmarking PII Leakage and Detection without Defense in Autonomous Web Agents
- Authenticity Debt and the Synthetic Content Threat Landscape: A Layered Framework for Trust, Provenance, and IP Governance in the Generative AI Era
- Lost in Delusion: Examining LLM Safety Under User Delusions and Distress
- Implicit Geographic Inference in LLM Medical Triage: Language-Driven Disparities in Emergency Recommendations
- KG-FairDiff: Knowledge Graph-Guided Prompt Refinement for Demographically Fair Text-to-Image Generation
- BenchEvolver: Frontier Task Synthesis via Solution-Centric Evolution
- GovAI-Pipe: A Layered AI Governance Pipeline for Citizen-Facing AI in Turkey's e-Government Gateway
- DrugClaw and DrugAudit: A Primary-Source-Grounded Agent and Authority-Aware Benchmark for Drug-Information Question Answering
- Dive into Ambiguity: A*-Inspired Multi-Agents Commonsense Obfuscation Attack on LLM Prompts
- LEGS: Fine-Tuning Teleop-Free VLAs for Humanoid Loco-manipulation in an Embodied Gaussian Splatting World
- ATLAS: Agentic Test-time Learning-to-Allocate Scaling
- Fair Finetuning Mitigates Distribution Inference Attacks
- SafeMCP: Proactive Power Regulation for LLM Agent Defense via Environment-Grounded Look-Ahead Reasoning
- AgentRedBench: Dynamic Redteaming and Integration-Aware Defense for LLM Agents over SaaS Integrations
- POIROT: Interrogating Agents for Failure Detection in Multi-Agent Systems
- Investigating and Alleviating Harm Amplification in LLM Interactions
- Food Noise & False Safety: A Systematic Evaluation of How LLMs Fail to Adapt to Eating Disorder Queries with Clinician Feedback
- Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modeling