Bloomberg Tech

Sarah Hiza on why Lockheed tests AI like an aircraft

Sarah Hiza· Senior Vice President of Technology and Strategic Innovation at Lockheed Martin
·~8 min·English·Bloomberg
LLMAI SafetyAI InfrastructureBusiness Strategy
TL;DR

Lockheed Martin's innovation chief explains how the F-35 maker runs on 55 models, tests mission AI the same way it flight-tests an aircraft, and now has an OpenAI team working beside its F-35 team on a sensor problem.

01Core Strategy

No Single Bet on a Frontier Lab

Lockheed treats the model layer as a horse race it refuses to call, so it runs on dozens of models at once rather than wiring the company to one vendor.

Fun fact, we have 55 different large language models that we're using ... To run Lockheed Martin.

— Sarah Hiza, Bloomberg Tech
Key Insight
Staying model-agnostic is a hedge against lock-in: with no clear winner among the frontier labs, Lockheed keeps its options open by running many models side by side instead of standardizing on one. The cost of that flexibility is operational, many models to integrate, secure, and keep current rather than a single stack.
02Safety

Flight-Test the AI Before It Flies

Asked how you stop a weapon-system agent from misbehaving, Hiza says Lockheed runs AI through the same testing gate it uses for an aircraft, probing the corners of the box before deployment.

We put it through rigorous testing, including corners of the box so that we understand its capability and to ensure the harnessing is sufficient.

— Sarah Hiza, Bloomberg Tech
Key Insight
Hiza's reassurance rests on an existing discipline: probe the AI's limits and confirm its harnessing is sufficient before fielding it. Two decades of machine-learning work, unblocked only recently when GPUs and TPUs broke the compute logjam, mean they are extending a known testing practice rather than improvising a new one for AI.
03Future of Warfare

The Win Is Making the Fleet Talk

Hiza sees AI reshaping warfare in two linked ways: crewed-uncrewed teaming, like a piloted F-35 flying with an uncrewed Vectis, and a shared data layer that finally lets today's separate systems talk to each other.

they all don't really talk to each other. So having the right data layer to have interoperability between all of those systems will be enabled by AI.

— Sarah Hiza, Bloomberg Tech
Key Insight
Both moves point the same way: the advantage comes from connection, not a single super-platform. Whether it is a crewed F-35 paired with an uncrewed Vectis or older systems linked through a shared data layer, the win is making separate pieces act as one.
04Talent

Mission Impact Is the Talent Draw

The AI work at Skunk Works is staffed by software, mechanical, and science engineers rather than only machine-learning specialists, drawn by the chance to see their work scaled to national impact.

we like to see our innovative results integrated but ultimately scaled and scaled at relevance so that we make an impact to the nation

— Sarah Hiza, Bloomberg Tech
Key Insight
The pitch to talent is deployment, not research for its own sake: the draw is watching a prototype become a fielded system. Working with venture companies and large commercial tech firms beyond its own walls also lets Lockheed import outside techniques without owning every lab.
05Emerging Tech

Quantum, Coming Out of the Lab

Hiza says quantum is near a breakthrough that moves it out of research, and Lockheed is aiming it at the sensing, secure-communications, and navigation problems specific to national security.

we're looking to use quantum for sensing, remote sensing, secure comms, and position navigation and timing.

— Sarah Hiza, Bloomberg Tech
Key Insight
Hiza names both quantum computing, which Lockheed expects to leverage much as it uses GPUs and TPUs, and a set of specific national-security uses: sensing, secure comms, and navigation. The demos planned for next year, some from space, are the near-term milestone she offers, not a fielded capability.
06The News

A Frontier Lab on the Flight Line

As Lockheed brings more advanced sensors onto the F-35, it has run into math and physics challenges, and Hiza says an OpenAI team is now working right beside the F-35 team to help solve them.

Their team is working right beside us. We are absolutely working right beside each other.

— Sarah Hiza, Bloomberg Tech
Key Insight
This is Hiza's beyond-the-walls approach made concrete: an outside lab pulled directly onto a live defense engineering problem. She confirms the two teams work side by side on the sensor math, but leaves the specific model and technical approach unstated.