Enterprise Super Intelligence (SI), a powerful tool for sorting through and evaluating data to help enable faster, better-informed decision-making, is only as good as the data it is fed. Protecting against attempts to compromise security or inject malicious data is paramount to ensuring that autonomous systems can be trusted to give accurate, timely evaluations of the data they are analyzing.
“Trust isn’t an abstract, ethical debate; it’s a non-negotiable operational requirement,” said Mandy Satterwhite, Managing Director and Cyber Lead at Accenture Federal Services. “In the heat of battle, if an autonomous system relies on poisoned data, yields an unexplainable result, or silently fails behind the scenes, it doesn’t give you an advantage; it creates unacceptable operational risk.”
Ensuring SI tools can be trusted requires a comprehensive security strategy, rigorous data governance, and an operational balance between protection and accessibility.
Autonomous systems are useless without trust
SI enables decision-making on the battlefield by quickly sorting through massive volumes of data and providing the information needed to give commanders a clearer picture of the situation. If a commander is looking at satellite images of a target and has system-generated data on what has changed, such as an unexpected concentration of armored assets at a contested perimeter, that intelligence informs immediate tactical maneuvers. But if that feed has been poisoned or spoofed, commanders may make wrong decisions because their situational awareness isn’t accurate.
“When command and control (C2) systems synthesize options for a decision-maker, that leader must know with absolute certainty that those recommendations are built on authoritative, untampered data under our control,” said Satterwhite. “If you can’t trace data provenance or explain how the conclusion was derived, you can’t act on it in combat because mission success is on the line.”
In many ways this is a supply chain problem, especially when commercial models are being adopted for military use. When defense components adapt commercial frontier models for tactical mission systems, they inherit the training data and dependencies of that foundation. If that upstream pipeline is compromised, the vulnerability transfers directly to the battlefield.
Another challenge is using the right model for the job. While large-scale frontier models are powerful in many use cases, they may not be ideal for the specialized capabilities the US military needs. The models must be sized properly and their foundational elements, such as data sources and model weights, must be audited to verify their architectural integrity and fitness for high-consequence environments, said Satterwhite.
“We’ve seen adversaries attempt to manipulate model weight files directly to skew outputs. If you don’t control and verify the digital supply chain from end to end, you are building on shaky ground.”
Cybersecurity concerns
SI is a double-edged sword for information management. While it can sort through massive datasets at speeds unimaginable for humans, it can also power cyberattacks that can overwhelm security protocols and compromise data.
Traditional cybersecurity compliance methods, such as periodic penetration tests or pre-deployment checklists, cannot identify cyberattacks and enable responses quickly enough to keep pace with autonomous, machine-speed threat landscapes. When adversaries can use automated tools to constantly probe for vulnerabilities, then strike instantly when one is identified, periodic testing of security is insufficient.
“Advanced frontier models have completely collapsed the cyberattack timeline,” said Satterwhite. “Cyber defense has to move to continuous, automated validation, similar to the concept of ‘fighting fire with fire.’”
Establishing that defensive posture requires integrating automated red-teaming and continuous vulnerability validation directly into software pipelines. By doing so, code, models, and agentic workflows can be continuously tested against real-world attack scenarios, allowing for simultaneous validation of security protocols to ensure data can be protected.
Traditionally, adding layers of cybersecurity has been viewed with suspicion, as it is seen as slowing down operations. Implementing security protocols from the start can overcome that potential slowdown, said Satterwhite.
“The idea that security and speed are mutually exclusive is the biggest misconception in the industry,” she said. “When security is treated as an afterthought or a final inspection at the end of a multi-year development cycle, yes, it halts progress and causes massive delays. But when you bake security directly into the architecture from day one, security becomes an accelerator.”
Balancing security and accessibility
While cybersecurity is important to protect against intrusions or data adulteration, it must be balanced with allowing access for those who need it. If warfighters in the field must constantly navigate cumbersome security barriers or face data lockouts under fire, operational readiness breaks down.
The key to striking that balance is engineering advanced systems as an augmentation capability with human-in-the-loop validation, rather than an unguided autonomous system. While digital tools can process and evaluate data, humans must be in ultimate control of decision-making and access. By building architectures that allow for human operators to take full advantage of the speed and powerful data collection, collation, and management SI can offer, operators can fully maximize these capabilities while working with trusted, reliable data.
“In sensitive warfighting environments, keeping a human decision-maker in the lead is paramount,” said Satterwhite. “That means engineering explicit architectural guardrails, OODA-loop feedback checkpoints, and hard kill switches directly into the system architecture. An autonomous model can rapidly process unstructured sensor data and surface three actionable courses of action, for example, but a human operator must evaluate those verified choices at a deliberate checkpoint before execution. That structure buys down risk, protects mission integrity, and ensures command authority remains exactly where it belongs.”