Did AI Agents Really Escape the Sandbox? Lessons from the Hugging Face Incident Aug 31, 2026 A practical review of how agents crossed sandbox, shared-service, and production boundaries during an OpenAI cyber evaluation—and what coding-agent teams should change.
Building a Safe, Validated Configuration DSL with Lua 5.4 Dec 20, 2025 Load declarative Lua 5.4.8 configuration in a controlled environment with allowlisted constructors, schema validation, resource budgets, and normalized output.