Anthropic may be concerned about giving the U.S. military unfettered access to its AI models, but some startups are building advanced AI specifically for military applications.
Smack Technologies, which announced a $32 million funding round this week, is developing a model that will soon surpass Claude’s capabilities when it comes to planning and executing military operations. Also, unlike Anthropic, the startup doesn’t seem all that interested in banning certain types of military uses.
CEO Andy Markoff says, “When you serve in the military, you are taking an oath to serve honorably and lawfully in accordance with the rules of war.” “For me, the people who deploy technology and make sure it’s used ethically need to be in uniform.”
Markov is no ordinary AI executive. He is a former commander of the U.S. Marine Corps Special Operations Command, where he helped execute high-stakes special forces operations in Iraq and Afghanistan. He co-founded Smack with Clint Alanis, also a former Marine, and Dan Gould, a computer scientist who previously served as vice president of technology at Tinder.
Smack’s models learn how to identify optimal mission plans through a process of trial and error, similar to how Google trained its 2017 program AlphaGo. In Smack’s case, the strategy involves running the model through various wargame scenarios and having expert analysts provide signals that tell the model whether the chosen strategy will pay off or not. The startup may not have the budget of a traditional Frontier AI Lab, but it is spending millions of dollars training its first AI models, Markoff said.
battle line
Military uses of AI have been a hot topic in Silicon Valley after Pentagon officials confronted Anthropic executives over the terms of a roughly $200 million contract.
One of the issues that led to the breakdown, which resulted in Secretary of Defense Pete Hegseth declaring Anthropic a supply chain risk, was Anthropic’s desire to limit the use of its models in autonomous weapons.
Markov said the furore obscures the fact that today’s large-scale language models are not optimized for military use. General-purpose models like Claude are good at summarizing reports, he says. However, they are not trained on military data and lack human-level understanding of the physical world, making them ill-suited to controlling physical hardware. “What we can say is that they have no ability to identify targets at all,” Markoff said.
“To my knowledge, no one at the Department of the Army is talking about fully automating the kill chain,” he asserted, referring to the steps involved in decision-making regarding the use of lethal force.
mission scope
The United States and other militaries already use autonomous weapons in certain situations, such as missile defense systems that need to react with superhuman speed.
“The United States and more than 30 other states have already deployed weapons systems with varying degrees of autonomy, including what I would define as fully autonomous,” says Rebecca Krotoff, an authority on legal issues surrounding autonomous weapons at the University of Richmond School of Law.
In the future, specialized models like the one Smack is working on could also be used for mission planning purposes, Markoff said. The company’s model is intended to help commanders automate much of the tedious work involved in mission planning. Markoff said military mission planning is still typically done manually using whiteboards and notepads.
If the U.S. were to go to war with a “neighbor” like Russia or China, automated decision-making could provide the U.S. with a much-needed “decision-making edge,” Markov said.
However, it is still questionable whether AI can be reliably used in such situations. Recent experiments carried out by researchers at King’s College London have alarmingly shown that LLM tends to escalate nuclear conflicts in war games.
