A speedboat bounces across the black ocean, racing through a moonless tropical night towards a small-deserted cove. Unseen by the drug runners aboard, their progress is being tracked by the thermal cameras of a sophisticated surveillance drone flying in lazy circles high overhead, the images beamed to the crew of a Coast Guard cutter cruising many miles offshore.

The cutter’s own radar can barely pick up the low-slung speedboat as it skips across the waves. But onboard, a Coast Guard commander has been tracking the drug runners’ progress since shortly after they loaded their illicit cargo and slipped out to sea: the type of boat they are using, combined with their course, and speed have led to them being flagged as suspect by sophisticated intelligence software that integrates data from a network of satellites, radars and drones. Using the thermal camera images from the drone and its own radar, the commander instructs the crew of two rigid inflatable Coast Guard boats that have been lying in wait around a nearby headland to gun their engines, relaying the course to steer as they race out to intercept the speedboat and its illicit cargo.

This vision of maritime surveillance, involving the layering of multiple kinds of sensor data and automated identification of potential “targets” using artificial intelligence, is the goal of Modern Intelligence, a small defense startup with big ambitions that is based in Austin. The company is an example of a bevy of new defense-related companies that are emerging around new technologies, such as drones and artificial intelligence, that hope to challenge the dominance and prevailing business models of the handful of “prime” defense contractors—such as Lockheed Martin, Boeing, and Raytheon—that have long dominated the industry. Since Russia’s invasion of Ukraine, these new defense startups are receiving a flood of attention from venture capitalists who have suddenly awoken to the sector’s importance.

“Now is the opportunity to build a generational defense company,” John Dulin, Modern Intelligence’s co-founder and chief executive, says.

Today, Modern Intelligence, which had been operating in “stealth mode” since its founding in late 2020, announced the completion of a $5 million seed funding round. The investment is being led by Geoff Lewis, the founder and managing partner of Bedrock Capital. Also participating in the funding round are Vine Ventures, a venture capital firm in New York; Air Street Capital, a London-based investment firm that backs startups using A.I.; and Contrary Capital, a venture firm that manages funds from co-founders and early employees of several well-known Silicon Valley successes, such as Airbnb, Facebook-parent Meta, Tesla, and SpaceX.

Modern Intelligence also said that it has been selected to test its maritime surveillance technology this summer at the U.S. Navy’s Naval Special Warfare’s Trident Specter exercises—which are designed to pilot emerging technologies that might be able to help U.S. Navy’s special operations forces. Modern has been developing its software partly with input from the U.S. military’s Southern Command and its Joint Interagency Task Force South, based in Key West, Fla., which is responsible for combating drug and people trafficking in Central and South America and the Caribbean.

Dulin says the company is trying to upend how military systems have traditionally been built and sold to the government. “Part of what is so messed up historically, is that everyone’s incentives have been to lock the customer into a vertical silo,” he says. As a result, data from one contractor’s sensor cannot easily be integrated with data from another’s, and every contractor is building its own A.I. systems to classify images. Modern Intelligence hopes to change that with a system built on simple open interfaces and application programming interfaces (APIs), just like most commercially available software.

This vision of open standards and full interoperability appealed to Jacqueline Tame, a former Deputy Director of the Pentagon’s Joint A.I. Center and former Congressional staffer, who agreed to join Modern’s advisory board. She told Fortune that when she was at the Department of Defense, she continually saw the U.S. military “get into these long-standing and protracted contracts by virtue of not understanding how the tech needs to be built to insure that doesn’t happen.” What the military needs, she says, especially when it comes to A.I. capabilities, is technology that is “agnostic to platform or infrastructure and can be interoperable with anything it needs to be interoperable with.” Modern, she says, is committed to building systems that can do this.

Also on Modern’s advisory board are Ellen Lord, the former CEO of defense contractor Textron and a former Undersecretary of Defense for Acquisitions and Sustainment; and Jason Yosinki, a co-founder of Uber’s A.I. research lab.

Besides developing its technology to work with sensors, equipment, and data from different defense contractors, Modern Intelligence also wants to deploy state-of-the-art machine learning methods that require less data to train and can deliver better analysis. For example, most A.I. algorithms that identify objects in images analyze them as flat, two-dimensional shapes. Modern Intelligence has developed a method to convert these to three-dimensional shapes, providing a far richer set of training data from the same source, Tristan Tager, Modern’s co-founder and chief scientist, says.

Besides requiring less data to train, the A.I. system is also more robust than others that are trained on large amounts of past data, but can’t cope when confronted with an object they haven’t encountered in training, according to Tager. “It can accurately identify and track a Chinese military vessel no one has seen before, that appears in no datasets, or where there were only one or two fuzzy images,” he says.

The company, Tager says, is also using methods for sensor fusion that are technically better than what many other defense contractors use. Most of these competitors, he says, don’t actually combine information from different sensors at all, but instead simply use the data feed from the sensor in which the A.I. system has the highest confidence at that particular moment in time, switching between sensors if that changes. Modern instead uses methods that can actually combine data, even when it arrives on vastly different timescales—such as the difference between video frame rates and still images from satellites that might fly over a particular location only once every few hours.

Tager says Modern eventually wants its maritime surveillance software, which it calls Cutlass, to provide analysts simple ways to set up high-level searches and alerts, perhaps even using natural language. For instance, he says, an analyst might simply type “alert me to any ship  within this operational area that is over 30 meters in length and exhibiting the following behavioral patterns,” or “show me all airbases in Western Russia where a T-160 aircraft is currently on the ground.”