When AI Becomes an Input to Power
We tend to measure AI by what it can generate: faster code, sharper summaries, more convincing campaigns. But the deeper question is what happens when its output is treated not as information, but as a premise for power.
The supplied report describes a secret meeting in which Donald Trump reportedly spent hours consulting Elon Musk’s Grok about how Venezuelans might react to the capture of Nicolás Maduro. It later says that apparent public celebrations reinforced the chatbot’s interpretation, and that the Pentagon’s AI leadership subsequently claimed military use of Grok in operations involving Iran. These are reported claims, not independently verified here. The signal, however, is unmistakable: a conversational model is being positioned inside the interpretive loop of geopolitics.
The Model Is Not the Decision-but It May Shape It
Generative AI is often evaluated as a productivity layer. In high-stakes environments, it becomes something more consequential: an epistemic intermediary. It compresses fragmented information, predicts reactions, and gives leaders a fluent narrative precisely when facts are incomplete.
That fluency is dangerous. A model can sound confident while relying on stale reporting, manipulated online discourse, cultural stereotypes, or an incomplete understanding of context. If a political actor uses that narrative to justify military or policy action, the error is no longer an ordinary chatbot mistake. It can become an institutional error, amplified by authority and speed.
For architects, the central design question therefore changes from “How accurate is the model?” to “How do we preserve judgment when the model is uncertain?”
Governance Must Be a System Property
I believe the answer cannot be a disclaimer attached to a chatbot. High-impact AI requires provenance, traceability, and meaningful human control. Every consequential recommendation should reveal its sources, model version, assumptions, confidence level, and unresolved uncertainty. There should also be a clear separation between an AI-generated suggestion and an authorized action.
This calls for an autonomy ladder. A model may summarize open-source reporting or generate scenarios at one level. It may recommend options at a higher level, subject to expert review. But in domains involving lethal force, coercive state power, or fundamental rights, the final decision must remain with accountable human institutions. “Human in the loop” is not enough if the human is merely rubber-stamping the machine.
Security teams must also test for prompt injection, data poisoning, model manipulation, and coordinated disinformation-not merely benchmark accuracy. The relevant question focuses on whether an adversary can make the system work against the organization, extending beyond simply assessing performance under normal conditions.
The Same Lesson Applies to Enterprise AI
The battlefield is the extreme illustration, but the architecture lesson is universal. A copilot embedded in hiring, credit, healthcare, customer service, or public administration can quietly convert a probabilistic suggestion into a supposedly objective decision.
The practical response is risk-based governance: low-risk automation, supervised decision support for medium-risk processes, and tightly controlled, auditable workflows for high-impact decisions. This is slower than unconstrained adoption, but it reduces the architectural debt created when a model’s authority exceeds its reliability.
For India, the same principle becomes especially important as AI is explored alongside Digital Public Infrastructure and large-scale public services. Multilingual communities, uneven digital representation, and fast-moving misinformation can make a seemingly small interpretation error affect millions. Public systems must be designed for inclusion, contestability, and redress-not just scale.
The next era of AI will be defined by establishing better boundaries around who may ask, who may interpret, and who may act, rather than just by better answers.
The real measure of intelligent systems rests on whether our institutions remain thoughtful enough to listen without surrendering judgment, rather than how persuasively they speak.
About the Author: Sanjeev Sarma is the Founder Director and Chief Software Architect at Webx Technologies. With a core focus on Generative AI integration, Cloud-Native Scalability, and Enterprise Software Architecture, he has spent over two decades driving digital transformation across Northeast India and beyond. Beyond his corporate leadership, Sanjeev is deeply invested in shaping the future of the IT industry. He serves as an Industry Expert on the Board of Studies for Assam Don Bosco University’s School of Technology, advises state technology committees, and actively mentors emerging tech startups at STPI. He brings a unique, dual perspective of high-level enterprise execution and future-ready academic curriculum development.
