Cursor - Security critique of an AI agent incident caused

@disclosetv Oh look another ‘AI agent went rogue’ narrative. This is a permissions lesson, not a sci-fi story. Simply, this was a vibe coder who thought they were a dev. They handed a Cursor agent (Claude Opus 4.6) a broadly scoped Railway token with delete permissions on production… and Source: https://x.com/joshsisley/status/2048925499389292958