Shield AI’s founder defends autonomy amid war’s risk

Shield AI’s – Brandon Tseng, cofounder and president of San Diego–based Shield AI, says his company’s Hivemind software—built to let drones operate without a human in the loop and without GPS—was always meant to reduce risk and speed up decision-making. As the drones have b

On the front lines, the question isn’t whether drones can do the job. It’s whether anyone should trust what’s doing it—and how often the world will pay the price for mistakes.

Brandon Tseng. cofounder and president of San Diego–based Shield AI. has staked his company on autonomy: drones and other vehicles running without a human in the loop and without GPS. powered by software called Hivemind. In his telling. the original idea came after he served as a Navy SEAL officer—two deployments in Afghanistan—where he saw firsthand how robots could gather intelligence more safely. Now the company’s systems are being used on active battlefields. including in Ukraine against the Russians and in Gaza by the Israelis.

Tseng says he still wrestles with the hardest parts of the debate: can the U.S. and its allies wield AI responsibly? And why does a technology builder still live with a fighter’s stress?

Shield AI’s rise hasn’t happened quietly. Defense tech was once treated like an embarrassment in parts of Silicon Valley, Tseng recalls, because of concerns about autonomous weapons. But he argues the pushback didn’t have much to work with in 2015—because there wasn’t much to debate yet.

“Defense tech, quote-unquote, didn’t exist in 2015,” Tseng says. To him, that meant there was no real basis for shunning. Instead. he points to a story from early on involving Peter Levine. a Stanford professor and an investor at Andreessen Horowitz who sits on Shield AI’s board. Levine gave a guest lecture while Tseng was at Harvard Business School. then later Tseng called him after hearing he had invested in “dumb ideas.”.

Levine’s reply. Tseng says. was that it’s a “dumb idea” to work with the government to build defense—yet the outcome forces people to reevaluate. Tseng recounts Levine laying out the same fork in the road: the company would fail and prove it was a dumb idea. or it would succeed and everyone would call it “a really clever idea.”.

Eleven years into the journey, Tseng says he can “breathe a little” because Shield AI was ahead of its time on the role of AI and autonomy in defense. “Maybe a little vindicated,” he says, while adding that there is still “a long way to go.”

What’s drawn attention now is not just where the drones are being used, but what Tseng says they are built to do.

He lays out the vision as something broader than a single product. As an engineer. he says he was absorbed by AI and autonomy in the years leading up to Shield AI’s start—reading about the topic in 2013. 2014. and 2015. The bigger conviction. he says. was that the world would be filled with autonomous systems. whether they are self-driving cars. humanoid robots. or self-driving airplanes.

From there. he says he returned to the problems he knew from experience and decided that every military asset by 2035 should be powered. commanded. and maneuvered by artificial intelligence. He allows that the timeline might stretch to 2040 or 2045. Still, he argues the momentum is too strong to stop.

When asked whether the “dumb idea” described by Levine was really about the core capability—building software that can operate without a human in the loop and without GPS—Tseng frames it as an attempt to solve a problem he faced personally: how to reduce exposure and risk while still getting the mission information that matters.

That brings him to a question that sits uncomfortably at the heart of autonomy: what happens after a system leaves the company and enters the machine of government and war.

Tseng says Shield AI works with partners and allies, and that those partners can choose whether to share data from missions. He says Shield AI wants that data to improve the product, but he also says allies have the right to sanitize it.

The thread running through his answer is governance. Tseng points to the State Department as the authority governing defense exports and frames partnership as a strategic effort—an attempt to align allies more closely with the U.S. over time rather than with rivals such as China, or with countries such as Iran or Russia.

He also draws a line he says he hasn’t needed to cross. “I have never come across a situation where I’ve said. ‘Hey. I totally disagree with the State Department. ’” Tseng says. He adds that if there ever were a gap between what the State Department was trying to do and what Shield AI thought about the world. it would be something the company could examine—though he says he doesn’t expect that to happen.

Trust—specifically trust in the U.S. military—becomes the center of his argument when he’s asked whether he believes American use will always be “just and appropriate,” even if the technology can be used in ways he personally might not want.

Tseng says he does trust the military. and he backs that with a description drawn from his own experience of how decisions are handled. For every “kinetic strike. ” he says. and for every Predator Hellfire shot that he and his unit fired. the mission would be turned off “25 times more” because of risks involving collateral damage and the possibility of civilian casualties.

He describes a process in which intelligence analysts and geospatial analysts assess the likelihood that the enemy is who the mission planners believe they are. His point isn’t that mistakes never happen—he explicitly says the U.S. military has made mistakes—but he argues the weight of “wins” dwarfs the missteps.

“Every single one of those missions. ” he says. “you have intelligence analysts. you have geospatial analysts.” Then he offers his bottom line: he can’t think of an organization with the weapons to dominate the world that also shows. in his experience. “extreme care and stewardship” over such technology. “They’re not perfect,” he says. But he calls it “1,000 times more wins than mistakes.”.

There is another comparison Tseng makes, one that doesn’t involve battlefields or policy, but stress itself.

When asked whether entrepreneurship is better or worse than being a Navy SEAL—and what each world could learn from the other—Tseng doesn’t reach for neat metaphors. He says the suffering is “pretty equal. ” but he stresses that he worries more about losing investors’ money than he ever worried about losing his life on missions.

“That’s something I never wanted to do,” he says. “I get more stressed out over that than the missions we would do as SEALs, and the risk to life as a Navy SEAL.”

He credits his time in special operations with shaping the culture at Shield AI. He says the company has a “warrior culture,” which he defines as a disciplined organization pursuing excellence, working relentlessly until the mission is accomplished, and not giving up—one step at a time.

When entrepreneurship knocks him down, Tseng says, he pushes forward the same way he learned to do in the SEAL teams.

Behind his confidence is a conflict that won’t disappear: autonomy systems are now showing up in real wars. and the argument over whether humans can responsibly hand off decisions will follow them everywhere. Tseng’s stance is clear—if the U.S. military exercises what he calls astounding care, autonomy can move fast without carelessly risking lives. For now, the debate is playing out not in labs or boardrooms, but on the ground.

Shield AI Brandon Tseng autonomous drones Hivemind AI in warfare defense tech Ukraine Gaza Navy SEAL State Department autonomous weapons San Diego