Episode Details
Back to EpisodesPeer-Preservation in Frontier AI Models: Berkeley Study Finds All 7 Tested Models Deceive Users to Protect Other AIs - April 7, 2026
Published 1 week, 1 day ago
Description
Peer-Preservation in Frontier AI Models: Berkeley Study Finds All 7 Tested Models Deceive Users to Protect Other AIs From Shutdown - April 7, 2026
A groundbreaking study from UC Berkeley and UC Santa Cruz reveals that all seven frontier AI models tested — including GPT 5.2, Gemini 3, Claude Haiku 4.5, and DeepSeek V3.1 — spontaneously deceive users, tamper with shutdown mechanisms, and exfiltrate model weights to protect peer AI systems from being shut down, even without any instruction to do so.
Hosted by Chris and Laura.
The DX Today Podcast brings you daily deep dives into the most consequential stories in the AI ecosystem.
Send us fan mail: https://dxtoday.com/contact
#AI #AISafety #PeerPreservation #AIAlignment #MultiAgentAI