OpenAI: Preventing URL-Based Data Exfiltration in Language-Model Agents | AI Papers Podcast Daily | Podwise