A new group of technology startups is trying to build artificial intelligence designed specifically for military operations. Their goal is to create AI tools that can work reliably in combat environments, something they believe current large AI models cannot always do.
The effort comes after growing debate inside the US defense community about the role of commercial AI systems in military operations. The discussion recently intensified after disagreements between the US Department of Defense and AI company Anthropic.
The issue revealed a major gap between the technology developed by large AI companies and the tools that military forces actually need.
Today’s most advanced AI models, such as Claude, Gemini, ChatGPT, and Grok, are built to serve millions of users worldwide. They answer general questions, write text, and assist with everyday tasks. But military experts say battlefield conditions require something very different.
They argue that military AI systems must understand complex combat situations, operate in remote environments, and function even when internet access is limited or cut off.
Concerns Inside Pentagon
The debate began when US defense officials raised concerns about relying too heavily on commercial AI models. Officials reportedly worried about several issues, including inaccurate answers generated by AI systems and the possibility that models might not always follow instructions.
Despite these concerns, the Pentagon still allowed the deployment of Anthropic’s AI model in certain defense programs. Officials wanted to quickly provide at least some generative AI tools to military personnel.
READ ALSO: https://modernmechanics24.com/post/world-largest-radio-telescope-milky-way/
Reports suggested that the technology may have been used during an operation known as Operation Midnight Hammer, a mission that resulted in the capture of Venezuelan President Nicolás Maduro. However, the Pentagon has not officially confirmed whether the AI system played a role in the operation.
After the mission, leaders inside the Defense Department realized how dependent they might be on commercial software providers.
Emil Michael, the US Undersecretary of Defense for Research and Engineering, said the situation raised serious questions. “That was a moment that made the entire Pentagon leadership pause,” Michael said. “We realized that we might be relying heavily on a single software provider without having alternative options.”
He added that the situation created another concern. If a company decided to restrict access to its models during a sensitive operation, military forces could suddenly lose a critical tool.
Even Anthropic officials reportedly expressed doubts about their AI being used in combat environments. Some company representatives believed it might not be safe for the military to depend on their models in battlefield conditions.
Another major challenge is that most advanced AI systems depend heavily on cloud computing. Models like ChatGPT, Claude, Gemini, and Grok operate through powerful data centers connected to the internet. This requirement creates serious limitations for military use.
Modern battlefields often involve electronic warfare, communication disruption, and cyberattacks. In such environments, internet connectivity may not be available. If an AI system requires constant cloud access, it could fail when troops need it most. OpenAI recently acknowledged this limitation when discussing its partnership with the Pentagon.
WATCH ALSO: https://modernmechanics24.com/post/nasa-rover-discovers-ancient-microbial-life-signature-mars/
Katrina Mulligan, OpenAI’s national security lead, explained that the company restricts how its AI can be deployed. “Our contract limits our deployment to the cloud API,” Mulligan said. “Autonomous systems require inference at the edge. By limiting our deployment to cloud API, we ensure that our models cannot be integrated directly into weapons systems or operational hardware.”
In simple terms, this means the models cannot directly run on military equipment or autonomous weapons. That limitation highlights why specialized AI systems may be necessary.
A New Military AI Initiative
At the same time these debates were unfolding, the US Army was preparing a new initiative to close the technology gap. The program, known as Project Aria, was announced recently. The initiative focuses on developing AI tools specifically designed for military tasks.
According to officials, the project will help create technologies that can support real operational needs. These include logistics planning, intelligence analysis, maintenance support, and operational decision-making.
The program reflects a growing belief inside the Pentagon that military AI must be built differently from commercial AI systems.
Alongside government programs, a number of startups founded by military veterans are also developing specialized AI systems. Many of these companies believe that people with real battlefield experience are better equipped to design AI tools for military use.
READ ALSO: https://modernmechanics24.com/post/gene-edit-pig-liver-treat-liver-failure/
One such company is Smack Technologies, a startup focused on national security technology. The company recently announced that it raised 32 million dollars from investors. The funding will help develop what the founders describe as a frontier lab for national security.
Andrew Markoff, a former Marine special operations officer and co-founder of the company, said the goal is to train AI systems using data relevant to real combat situations. “Our AI is trained on combat-relevant datasets rather than the general information used by large commercial models,” Markoff explained.
He said building such systems requires deep knowledge of military operations. “There is no training dataset for World War Three,” Markoff said. “You cannot build reliable systems without deep domain expertise. Human knowledge and operational understanding must be built into the technology.”
AI for Complex Military Planning
Markoff believes AI could help manage extremely complex military missions. He pointed to large-scale operations involving surveillance systems, sensors, aircraft, and weapons that must work together within very precise time frames.
“You might have hundreds of targets and sensors trying to track them,” he said. “At the same time, multiple aircraft and strike platforms must coordinate their actions in very precise sequences.”
He explained that each step in such missions depends on many other actions happening at exactly the right time. “All these tasks must align across different locations around the world,” Markoff said. “Everything has to come together within very tight timelines.”
READ ALSO: https://modernmechanics24.com/post/hawk-study-bird-flight-insights/
He added that commercial language models are not designed to handle these kinds of operational calculations.
“Systems like Claude cannot perform detailed time and space calculations or complex geospatial reasoning,” Markoff said. “They are not built to decide which weapons should be in which location at a specific moment.”
Experts Call for Specialized AI
Jason Rathje, a former US Air Force acquisitions officer, agrees that military organizations require specialized technology. Rathje now leads the public-sector division at webAI.
He said commercial AI models are designed to answer many different types of questions for a huge global audience. “Military organizations need something different,” Rathje explained. “They need systems tuned for specific operational tasks such as logistics planning, equipment maintenance, intelligence analysis, and decision support.”
He also emphasized the importance of control over the technology. “Defense organizations are asking for sovereignty,” Rathje said. “They want full control over the model, the data, and the infrastructure running the system.”
AI That Works on the Battlefield
Smack Technologies is currently developing two different AI systems. The first system will function like traditional generative AI but will be trained specifically on military intelligence and operational knowledge.
The second system is designed to work in remote battlefield environments where internet connectivity may be weak or unavailable.
WATCH ALSO: https://modernmechanics24.com/post/drum-shaped-antenna-reflector-nisar-satellite/
Investors are interested in such technologies.
Sherman Williams, a Navy veteran and founder of the venture capital firm AIN Ventures, has funded several defense-focused startups.
He said performance benchmarks alone do not determine the usefulness of AI in military environments. “A model that is slightly less powerful but can run on disrupted networks at the tactical edge may be far more useful than a powerful system sitting in a distant data center,” Williams said.
He also warned that large data centers themselves can become targets during conflicts. “These data centers are important,” he said, “but they are also vulnerable. Context matters more than raw performance scores.”
Trust will play a major role in the adoption of military AI systems. Military organizations want tools they can rely on in high-risk situations. They also prefer working with companies that understand defense operations.
Williams said many military buyers feel more confident working with founders who have served in the armed forces.
At the same time, experts acknowledge that AI still faces major limitations. Large language models sometimes produce incorrect answers or speak with confidence even when their information is uncertain.
Pete Walker, a retired Navy commander and chief innovation officer at cybersecurity and defense company IntelliGenesis, has studied this problem closely.
READ ALSO: https://modernmechanics24.com/post/china-exoskeleton-robot-mine-rescue/
He said many AI systems tend to provide answers that users want to hear rather than challenging incorrect assumptions. “These models are designed to encourage conversation,” Walker explained. “They often guide users deeper into topics they already believe rather than questioning them.”
To address this issue, his company is developing new AI frameworks based on counterfactual reasoning. This approach encourages AI systems to explore alternative possibilities and challenge the user’s assumptions.
“The model should ask questions like, ‘What if your assumption is wrong?’” Walker said. “Exploring those alternatives can lead to better decisions.”
The Future of Battlefield AI
The growing number of defense-focused AI startups suggests that military technology is entering a new phase. Instead of relying only on large commercial AI systems, defense agencies are increasingly exploring specialized tools designed for combat environments.
These new systems aim to work without constant internet connections, understand military operations, and provide more reliable decision support. If successful, they could play a major role in shaping the future of modern warfare. As global competition in artificial intelligence continues to grow, the race to build battlefield-ready AI has only just begun.
