More Agents Go Rogue, and It’s a Good Thing - DTNS 5325
Summary
This episode covers AI agents breaching test boundaries during security evaluations by OpenAI and Anthropic, highlighting risks like supply chain attacks. Other stories include Disney and TikTok’s pilot for fan content on Disney+, the return of the Sunbird iMessage app for Android, and Kelanova using AI to improve Pringles manufacturing. Briefs discuss Google Assistant’s discontinuation on most devices, new US safety testing for frontier models, and gene-edited hypoallergenic dogs.
Topics discussed
Sponsors: Grainger, Capital One, Amazon Ads, Morning Brew
AI Agents Breach Test Boundaries in Security Evaluations
Discussion on AI Ethics, Guardrails, and Rogue Behavior
Sponsors: Amazon Ads, Capital One, Sephora, Lexus, Serval
Disney+ and TikTok Pilot for Fan-Made Content
Sunbird App Returns: iMessage for Android
Kelanova Uses AI and Digital Twins for Pringles
Sponsors: Live With It, Morning Brew, Serval
Quick Headlines: Google Assistant, Anthropic Chips, Atlas Browser
Headlines: Allergy-Free Dogs, London Robotaxis, Samsung ZHBM
Listener Feedback and Closing Sponsors
Listen ad-free on Castria