BAE Systems Put AI Role Players in Its War Simulator—The ‘Enemy’ Now Never Sleeps
BAE Systems' Mimesis maritime simulator now uses a large language model as a role player, trained with call signs, responses, and doctrine. The enemy never sleeps, never gets tired, and costs less.
- BAE Systems deployed AI role players in its Mimesis maritime battle simulator, replacing costly human actors with a large language model.
- The system was trained by telling the AI: this is your role, call sign, responses, and allowed actions.
- The same dynamic is hitting every domain where human role players once scaled infinitely.
Training a warfare team has always required experienced role players who know when to escalate, when to hold back, and when to throw a realistic wrench into the scenario. BAE Systems has concluded that you can replace most of them. At the International Training Technology Exhibition and Conference (ITEC 26) held in London in April, the defense contractor unveiled an AI-driven role player integrated into its Mimesis maritime battle simulator—the first time the system has been demonstrated publicly.
Richard Goldstone, BAE Systems’ business development lead for maritime simulation and training, described Mimesis as a synthetic representation of the maritime warfare environment that can model the movement and behavior of ships, submarines, and other platforms. It digitally replicates combat management systems, sensors, and effectors—and can replicate threats and stimulate operational software. The problem it solves: human role players are expensive, scarce, and limited by what they can individually absorb.
“To get the correct responses we told it: this is the role you’re playing, this is your call sign, this is what you respond to, these are the actions that you’re allowed to take in response to this input,” Goldstone told Janes. The AI was built by training a large language model on those parameters. The result is a role player that never gets tired, never deviates from doctrine, and can run the same scenario 50 times in a row with consistent, evaluable behavior.
The Cost Equation Is Changing Fast
The traditional model of distributed training—where participants role-play different elements of a scenario within a larger exercise—has been a workaround for the scarcity of human role players. But it has limits. Distributed setups still require a trained human to play each significant role. An AI role player removes that constraint entirely, enabling scenarios with multiple simultaneous threats, complex coordinated attacks, or adversary behavior patterns that would require a dozen trained actors to simulate realistically. BAE is positioning this as a cost reduction story as much as a capability enhancement—the same scenario, no additional personnel costs, no scheduling conflicts, no fatigue.
The maritime context matters. Naval exercises involve long-duration scenarios where the quality of adversary behavior directly affects what crews learn. A passive or inconsistent role player trains bad instincts. An AI that can sustain a multi-hour coordinated submarine evasion scenario at full fidelity is qualitatively different from what most naval training programs can afford to run regularly. The technology also means the scenario library is limited only by the training data—no physical infrastructure needed beyond the simulator itself.
Commercial-off-the-shelf AI for defense applications is not unique to BAE. The Pentagon’s Project Maven has been deploying AI for battlefield analysis, and the U.S. Department of Defense has been expanding AI usage across military branches. But in the simulation domain, where the AI doesn’t need to make life-and-death decisions—just consistent, doctrine-compliant ones—COTS AI is arriving faster and with fewer regulatory hurdles than autonomous weapons systems. The Locked Shields 2026 exercise, involving 41 nations practicing cyber resilience, reflects the same dynamic across a different domain: AI tools are making traditional training exercises denser and more realistic, faster than doctrine can adapt.
The next step for Mimesis appears to be expanding the range of roles the AI can play—currently focused on maritime warfare, but the architecture is not domain-specific. Whether BAE can replicate the approach for land or air training scenarios will determine whether this is a maritime niche product or the leading edge of a broader shift in how militaries conduct realistic training at scale.
