We were shown a demonstration of the AI system that powers Anduril’s vision for war
Right away, humans are also required to translate between systems from different manufacturers. A soldier may have to rotate a camera manually to scan a base to see if a drone is present. If so, they will have to send the information to another soldier who controls the weapon. They could use a messenger app similar to AOL instant messaging for this purpose. This takes time. It’s something the Pentagon is attempting to solve through its Joint All-Domain Command and Control plan, among other initiatives.
“For a long time, we’ve known that our military systems don’t interoperate,” says Chris Brose, former staff director of the Senate Armed Services Committee and principal adviser to Senator John McCain, who now works as Anduril’s chief strategy officer. His work has largely been to convince Congress and the Pentagon of the importance of software problems in the defense budget, just like jets and aircraft carrier. Anduril spent $1.6 million lobbying in the last year according to Open Secrets. It has many ties with incoming Trump administration. Palmer Luckey, founder of Anduril, has been a supporter and donor of Trump for a long time. JD Vance, who worked at Revolution Venture Capital, spearheaded an investment into Anduril. Defense hardware is also plagued by a connectivity issue. Tom Keane is a senior vice-president in Anduril’s connected warfare division. He walked me through an example that was taken from the civilian sector. When you receive a message on your phone, it will appear when you turn it back on. It’s preserved. Keane explains, “This functionality, of which we don’t think at all,” “it does not really exist” when it comes to the design of many defense hardware systems. In difficult military networks, data and communications are easily lost. Anduril claims that its system stores data locally. An AI data treasure-trove
The Pentagon’s push to create more AI-connected systems for the military may spark the biggest data collection project it has ever undertaken. Companies like Anduril, Palantir and others have ambitious plans.
“Exabytes worth of data for AI training, inferencing and analysis are disappearing,” Anduril stated on December 6 when it announced that it would work with Palantir in order to compile the data collected by Lattice. This includes highly classified information. Anduril’s partnership with OpenAI announced on 4 December will greatly boost its model-building efforts by allowing it to train on a larger collection of data from all these sensors. Palantir offered its AI tools earlier this year to help the Pentagon reimagine their categorization and management of classified data. Palmer Luckey, the founder of Anduril, told me during an October interview that there was not a lot of data on classified topics or understanding of weapon systems to train AI models. He may have foreshadowed what Anduril now is building. Even if the data is already collected by the military, AI will make it more useful. Emelia Probasco is a senior fellow with the Center for Security and Emerging Technology of Georgetown University. She wrote an email saying that “what is new” is the ability for the Defense department to use data in new and innovative ways. “More data and ability to process it could support great accuracy and precision as well as faster information processing.”
The sum total of these developments might be that AI models are brought more directly into military decision-making, rather than just surfacing information. This idea has been scrutinized, as when Israel was discovered to be using advanced AI models last year to process intelligence data to generate lists of target. Human Rights Watch stated in a recent report that these tools “rely upon faulty data and inaccurate approximations”.
“I think we’re already on the path to integrating AI into decision-making, including generative AI,” says Probasco who wrote a recent report on one of these cases. She studied a 2023 system called Maven Smart System that allows users to “access sensors data from various sources
and apply computer vision algorithms in order to help soldiers select military targets.”
