Daily Research Brief 2026-05-27

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

The paper thread today: AI safety as effective controllability, in-context continual unlearning (ICCU), scaling the harness rather than the model for long-horizon agents, and fine-grained credit assignment for LLM training.

1. Latest arXiv Papers

1. Position: AI Safety Requires Effective Controllability — https://arxiv.org/abs/2605.27117

2. ICCU: In-Context Continual Unlearning via Pattern-Induced Forgetting — https://arxiv.org/abs/2605.27138

3. Advancing Mathematics Research with AI-Driven Formal Proof Assistance — https://arxiv.org/abs/2605.22763

4. From Model Scaling to System Scaling: Scaling the Harness for Long-Horizon Agents

5. MobileGym: A Verifiable and Highly Parallel Simulation Environment for Mobile Agents

6. SafeDiffusion-R1: A Continual Learning Framework for Safe Diffusion Models

7. SkillEvolver: Self-Evolving Skill System for AI Agents

8. Fine-Grained Credit Assignment for LLM Training

9. LingBot-VA: A Causal World Model for Robot Simulation and Manipulation — https://arxiv.org/abs/2601.21998

10. Squeezing Capacity from Multimodal Large Language Models