-[SECURE]- reads a system before the system ships — control flow, trust boundaries, the assumptions a caller is making — the way an auditor reads, not the way a linter does. The long-term product is a static detection agent. What ships today is stage 1 of that work, run as a guided service.

Researchers drive the tooling by hand. The same corpus that trains -[CONTAIN]- is what the review is tuned on. The agent that runs this without a human in the loop is what comes next.