MIT Tech Review bundles six military AI decision stories into an eBook

MIT Technology Review has put a spotlight on the most consequential question in military AI right now: when systems stop advising and start deciding. The outlet’s subscriber-only eBook, authored by James O’Donnell, is a package of six stories about militaries using AI models to make decisions. The collection is updated to reflect recent developments and includes reporting originally published between April 11, 2025, and April 21, 2026.

If you are an executive, board member, or investor trying to understand where the AI money and risk are headed, this eBook functions like a field guide. It explicitly frames AI as moving into the “decision” layer, not just the “analysis” layer, and it organizes the themes around what militaries are building, testing, and planning. The related stories list within the eBook landing material makes the direction even clearer: “The new war room,” “Generative AI is learning to spy for the US military,” “Phase two of military AI has arrived,” and “A defense official reveals how AI chatbots could be used for targeting decisions.”

That “targeting decisions” phrase matters because it signals the practical endgame of most military AI programs. There is a difference between a tool that summarizes information and a system that influences which target gets prioritized, which action gets approved, and how quickly a human operator can respond. Even without reading each individual piece, the eBook’s curation tells you the newsroom sees a shift from earlier experimentation to operational consideration. The collection also points to how militaries are increasingly treating AI as part of the workflow, what many companies would call a “decision stack” rather than a standalone product.

There is also a technology and business sub-plot hiding in plain sight. The eBook is packaged as six stories, and the related items expand the scope from capabilities to procurement and data strategy. One listed story is “The Pentagon is planning for AI companies to train on classified data, defense official says.” That is an inflection point for AI vendors: model access is no longer limited to public datasets and sandboxed testing. It raises hard questions for compliance, custody, and lifecycle management of sensitive data, because training on classified data is not just a technical upgrade. It is an operational burden, a contract structure change, and a governance challenge.

Executives should also pay attention to how this kind of work gets sold internally and externally. Military AI programs live at the intersection of national security priorities, procurement constraints, and institutional risk tolerance. When a publication bundles multiple stories about decision-making and targeting, it is essentially documenting how stakeholders are aligning around specific uses of AI. For boards and leadership teams at AI companies, that alignment can change the sales cycle. It can also change the liability posture, because the moment AI influences decisions, the organization using it will want clarity on accountability, auditability, and performance under stress.

Finally, there is the strategic ripple effect across the whole ecosystem of defense tech. One related story in the eBook listing is “How AI is turning the Iran conflict into theater.” Another is “Phase two of military AI has arrived.” Those two headlines together suggest a second-order reality: AI is not only improving military decision support. It is shaping the information environment around conflicts, which can accelerate demand for AI tools that can operate at speed, interpret signals, and help coordinate actions. Even if the exact mechanics differ by scenario, the direction is consistent with the eBook’s central theme, AI is moving deeper into how militaries interpret situations and act on them.

For decision-makers evaluating where to allocate budgets, talent, and product roadmaps, the takeaway is simple. If militaries are planning for AI companies to train on classified data and considering AI chatbots for targeting decisions, then the center of gravity shifts toward responsible deployment, secure data pipelines, and systems designed for real decision contexts. That is a different game than building a model for general-purpose predictions. And the sooner your organization understands the decision layer, not just the model layer, the better positioned you will be when procurement, regulation, and competition catch up to where the technology is already going.