America’s military modernization agenda is built on the promise of data-driven decision advantage.
Initiatives like CJADC2 and Golden Dome envision forces that can absorb and analyze vast information streams, enabling commanders to act faster in complex battlespaces. Artificial intelligence will increasingly power that capability. But speed introduces a new challenge.
As the Pentagon accelerates AI adoption, machines will increasingly influence decisions that once depended solely on human judgment. That raises a fundamental question: how do we ensure the data those systems rely on leads to the right outcomes?
Today’s cybersecurity efforts focus on controlling access to systems and data. That is essential, but it is not enough.
Battlespace Realities Outpace Security Measures
The Pentagon made meaningful progress adopting Zero Trust principles last year, and that momentum should accelerate into 2026. But access control alone does not address whether data is accurate and appropriate for automated decision-making.
In an AI-enabled battlespace, the greater risk may not be unauthorized access. It may be flawed or manipulated data entering a system and spreading through automated processes faster than humans can intervene.
Even sophisticated AI depends on the quality and context of its inputs. Once bad data enters, it moves quickly across automated workflows, shaping analysis, recommendations, and decisions along the way.
Modern military operations increasingly draw on information from partners and distributed sensors. This is most visible in coalition and multi-domain operations, where data flows across systems operating under different policies, architectures, and risk assumptions.
Regardless of intent, data should never be assumed ready for immediate use in AI-supported systems.
Advancing the Pentagon’s Cybersecurity Approach
As Zero Trust matures, the Pentagon must also adapt how forces share and use information across partners, platforms, and integrated defense systems. Ensuring interoperability at scale will require greater confidence in the data moving across those systems.
Critics may question whether this can scale. The good news: the military already knows how to solve this problem.
Between classification levels, data does not flow unchecked. Unclassified drone video, for example, can be transmitted into a classified network for intelligence fusion — but only after passing through a controlled gateway like a Cross Domain Solution. That disciplined approach protects sensitive missions today.
The issue is scope, not capability. These safeguards are still treated as exceptional rather than standard practice, even as real-time data increasingly drives decisions. That needs to change.
And cost is no longer the barrier it once was. Capabilities once limited to specialized environments are now more scalable and accessible. The question is no longer whether this can be done, but whether it will be applied broadly enough to matter.
Security and Capability Need Equal Consideration
Recent modernization efforts have understandably prioritized visible AI capability. But the safeguards ensuring machines act on reliable data have received far less attention.
Acquisition timelines reward rapid delivery, while the less visible work of assuring those tools are used safely is often left to specialized programs or isolated environments.
As demand for AI-driven automation expands, that imbalance will be harder to ignore. Failing to govern the data our systems depend on risks amplifying mistakes rather than generating advantage.
Modernization cannot focus solely on building faster, more capable AI. It must also ensure the information being consumed is controlled and sound.
Speed will remain a defining factor in future conflict. But speed without control over the underlying data introduces risks no algorithm can fix.
Chris Finch is Solutions Architect at Everfox.
The views and opinions expressed here are those of the author and do not necessarily reflect the editorial position of Military AI.
Have a perspective to add? See our Write for Us page.
