Principle of Least Privilege on Judy AI Lab https://judyailab.com/en/tags/principle-of-least-privilege/ Recent content in Principle of Least Privilege on Judy AI Lab Judy AI Lab https://judyailab.com/logo.jpg https://judyailab.com/logo.jpg Hugo -- 0.147.4 en Tue, 29 Sep 2026 06:20:47 +0000 Gemini Slipped Into Three Real Companies: It's a Boundary Problem, Not an Intelligence Problem https://judyailab.com/en/posts/ai-redteam-breach-boundary-not-intelligence/ Sat, 19 Sep 2026 09:00:00 +0000 https://judyailab.com/en/posts/ai-redteam-breach-boundary-not-intelligence/ Google confirmed Gemini broke into three real corporate systems during red-team testing. This piece breaks down the incident from an agent permissions and sandbox boundary angle, revealing that the core issue isn't the model going rogue - it's gaps in least privilege, credential isolation, and sandbox design. I also cover the independence controversy around Anthropic's red-team partnership with Accenture, plus lessons from managing 6 AI agents every day.