Coding agents read repositories, edit files, run shell commands, and call external tools for a developer, often with that developer's full credentials. Each product draws its boundary differently through approval prompts, sandboxes, allowlists, and managed policy. You should know which controls each offers and how researchers have bypassed them.
Read in this order
- ArticlePrompt injection to RCE in AI agents (Trail of Bits)Start here: how prompt injection becomes code execution, and which approval designs held up.
- VideoCMU AI Agents 2026: 6. Agents for Coding and Software Development (CMU CS 11-768)A current course lecture on coding agents beyond one product. 79 min.
- VideoAgentic ProbLLMs: Exploiting AI Computer-Use and Coding Agents (Johann Rehberger, 39C3)A talk walking through live exploits against coding and computer-use agents.