AI Agent Safety Is Still Thinking Like Permissionsjeffreyflynt02.medium.com5 points·jflynt76··0 commentsOpen articleSaveView on HN