Daily Research Brief 2026-05-11

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

SpaceX signed with Anthropic for Colossus-1 datacenter access; Ollama disclosed high-severity CVE-2026-7482; RPCS3 banned automated AI code submissions; the SGLang team raised a $100M seed round. On the paper side: safety certification as classification, VibeServe bespoke LLM services, and World-R1 3D-constrained video generation.

1. Latest arXiv Papers

1. Safety Certification is Classification — https://arxiv.org/abs/2605.06087

2. VibeServe: Can AI Agents Build Bespoke LLM Services? — https://arxiv.org/abs/2605.06068

3. Visual Fingerprints for LLM Generation Comparison — https://arxiv.org/abs/2605.06054

4. Ex Ante Evaluation of AI-Induced Idea Diversity — https://arxiv.org/abs/2605.06540

5. MedHorizon: Towards Long-Context Medical Video Understanding — https://arxiv.org/abs/2605.06537

6. PACZero: PAC-Private Fine-Tuning of Language Models — https://arxiv.org/abs/2605.06505

7. World-R1: Reinforcing 3D Constraints for Text-to-Video Generation — https://arxiv.org/abs/2604.24764

8. BARD-VL: New SOTA in Multimodal Diffusion Models — https://arxiv.org/abs/2604.16514