Lockheed puts military AI through RAGE review before launch

Lockheed Martin positions itself as a producer rather than a consumer of AI, and the LAIC's mandate is to do the hard work once and share it across the enterprise. Mike Harasimowicz calls this "build once, use many": an optimization developed for sea-state conditions can be adapted for land and air and applied across guided missile cruisers, advanced fighters and ground-based missile defense systems. The LAIC also keeps an AI inventory so new capabilities complement rather than duplicate what already exists, which Lockheed says guards against redundancy and wasted budgets.

The RAGE framework reflects a red-teaming heritage that Lockheed traces to Air Force Red Flag exercises and to its own cyber kill chain work. Its tollgates ask why a system is built, whether the architecture and computing capacity are right, whether it is strong enough, and how performance will be monitored in production. Harasimowicz describes the stance bluntly: trust that has not been challenged is not trust but hope, and the goal is to make Lockheed a harder target than an adversary wants to engage.

Lockheed also flags internal risks: automation bias, where operators stop questioning outputs, and data poisoning, where adversaries corrupt training data. Its answer includes regular review — Harasimowicz compares an AI model to an airplane coming in for maintenance — and a principle that humans stay accountable for decisions even when AI-enabled systems are involved. Lockheed says that experience gives it standing to help set standards at company, national and international levels, though the piece sets no timeline for that.