Nigel Robinson

SIGIL: Construction-Level Security for Agent-Authored Code

Large language model (LLM) agents increasingly write, compile, execute, and revise software inside systems that also hold files, credentials, and network authority. In that setting, security cannot depend on human review, external sandboxes, or post-hoc filtering alone: those mechanisms treat generated code as an opaqu...