Autonomous systems are being rapidly integrated on active battlefields, and the AI models driving them are advancing in ways even their developers cannot fully predict. That pairing defines the central technical condition of AI warfare.
The specific constraint is predictability. As frontier AI models advance, they develop in ways their builders did not anticipate. That property does not change when these systems are moved from a controlled environment to a combat one. The consequences attached to uncharted model behavior simply shift in kind.
The integration underway involves autonomous systems, meaning systems that act without human confirmation at each step, a distinct deployment regime from AI used as an advisory layer behind a human decision. Operating at that level of autonomy while the underlying models remain difficult to forecast is the problem now being inherited by the battlefield.
The models' own creators cannot fully predict how they will advance. The systems built on them are already in the field.