Back in the Cold War days, a perceived missile gap between the USA and the Soviet Union was a driving theme behind U.S. military growth. But while that missile gap proved to be largely illusory, an AI gap between the U.S. military and the rest of the world’s armed forces is no hallucination.
If there was a single doubt about AI’s military prowess in the war with Iran, it centered around the tragic missile strike on an Iranian elementary school which killed as many as 175, mostly children. Artificial intelligence emerged as a prime suspect but the consensus is that this targeting error is due to human error, according to Semafor. Human reviewers failed to recognize subtle changes in satellite imagery while also missing publicly available information concerning the school.
The Pentagon’s AI targeting scheme is basically an artificial mind meld between Palantir’s Maven and Anthropic’s Claude AIs. Maven identifies potential targets from satellites and other sources while Claude helps planners sort the information to decide on which targets to prioritize. The emphasis is on speed. AI compresses what the military calls the “kill chain” from days to hours. But while the Pentagon’s use of AI is impressive, its ongoing dispute with Anthropic raises some concerns that the Pentagon is shooting itself in the artificial foot. The Pentagon has ordered all military units to remove AI products developed by Anthropic within 180 days after the company was designated a “supply chain risk,” an unprecedented move against an American company that will have its day in court.
Also casting shade on the use of AI in military campaigns is a new poll by YouGov for the Economist that shows 44% of Americans agree that AI companies should be able to put restrictions on how the military uses its technology (32% are unsure). The Anthropic dispute revolves around that crucial question. The American public also voted with its feet as uninstallations of ChatGPT surged 295% when it stepped in to offer the military an alternative to Anthropic. The Pentagon wants unfettered use of AI while Anthropic refuses to remove safeguards that prevent its use for autonomous targeting and the conducting surveillance operations against U.S. citizens. Anthropic CEO Dario Amodei warns that if AI is misused, it could enable a central command or even a single leader to direct powerful weapons networks.
There is no doubt, however, that the speed at which the Pentagon has integrated AI into military operations has stunned both allies and potential adversaries alike. AI enabled Operation Epic Fury to strike over 1,000 targets in the first 24 hours while also working in concert with the Israeli military to eliminate senior Iranian officials, including most famously Iran’s Supreme Leader. While much of AI’s military role is classified, it’s clear that AI dramatically accelerated the processing of signals intelligence, imagery and open-source data into actionable targeting packages. AI also likely was used with Lucas one-way attack drones to provide near-autonomous navigation as well as target guidance and recognition.
The perception of a military AI gap with the U.S. is widening. China emerged as a spokesperson of sort, warning against a Terminator-style future should the U.S. grant AI the ability to decide life and death on the battlefield. Multiple news reports painted an internal picture more alarming, with Chinese advisors warning the country’s leadership that Beijing “risks making historical mistakes” by emphasizing its AI capabilities primarily in the entertainment and civilian sectors and not accelerating AI military development. Militaries ranging from South Korea to Finland also are now emphasizing AI programs.
The race to catch up is on but the Pentagon isn’t waiting around. The U.S. Army, for example, is accelerating the use of AI across its ranks with the launching of Project Aria to push AI tools into the hands of soldiers faster. A key element in this initiative is the development of a Global Unified Network that gets AI decision tools into the hands of frontline troops. The Army also is using AI for better logistics planning: the Gallatin Navigator analyzes mission data to predict supply demands to quickly generate multiple logistics solutions.
Meanwhile, some are betting the dispute between the Pentagon and Anthropic will ultimately lead to a divorce between the military and AI systems with origins in the civilian sector. One is a London-based startup called Hadeon with investment backing from Booz Allen Hamilton that aims to scale up large military operations in a simulated environment. Another is a Californian startup called Smack Technologies that has raised $32 million to build AI models with data solely derived from the military domain to speed the decision making.
A military-specific AI doesn’t need to include fictional military data scrapped from a Tom Clancy novel like The Hunt For Red October, say proponents of this approach, a glib acknowledgement that commercial AI systems designed to answer questions from millions of users is full of data the military doesn’t need while still unable to do tasks like determining which munitions need to be where at specific times. The Pentagon also worries that commercial usage restrictions could pop up in the middle of an operation. The Pentagon also wants AI systems that don’t need a connection to the cloud, a feature of commercial AI models that is deemed a vulnerability by the military.
In the short term, however, the Pentagon’s use of AI has proven to be overwhelming in the war with Iran. At the same time, that success still comes with a dose of tension. And while there may be larger questions about the use of AI in war, that horse has left the American barn. Now everyone is trying to catch it.
