Silicon Valley Debates the Role of AI in Autonomous Weapons & Warfare

In late September, Shield AI co-founder Brandon Tseng expressed confidence that weapons in the U.S. would never be fully autonomous, reassuring that AI algorithms would not be making the final decision to take a life. “Congress doesn’t want that. No one wants that,” Tseng told TechCrunch. However, his statement was quickly followed by a contrasting viewpoint.

Five days later, Anduril co-founder Palmer Luckey voiced skepticism over arguments against autonomous weapons during a talk at Pepperdine University. Luckey questioned the moral high ground of rejecting AI decision-making in warfare, asking, “Where’s the moral high ground in a landmine that can’t tell the difference between a school bus full of kids and a Russian tank?”

A spokesperson for Anduril later clarified that Luckey’s comment didn’t advocate for autonomous weapons but raised concerns about the dangers of “bad AI” in the wrong hands. This debate exposes the growing tension between the defense tech industry’s potential and the ethical implications of AI in warfare.

Cautious Perspectives in Silicon Valley

Other industry leaders remain more cautious. Trae Stephens, another co-founder of Anduril, highlighted that AI technology could be a tool to support better human decision-making in lethal situations. While Stephens emphasized the importance of accountability, he didn’t insist that a human always needs to make the final call, just that someone must remain responsible for decisions involving lethal outcomes.

Ambiguity in U.S. Policy on Autonomous Weapons

The U.S. government’s position on fully autonomous weapons remains unclear. While the military does not currently use them, there are no outright bans preventing companies from developing or selling autonomous systems. Last year, updated guidelines for AI safety in military applications were introduced, requiring approval from top military officials for new autonomous weapons. However, these guidelines are voluntary, and U.S. officials have expressed reluctance to impose binding bans.

Pushing the Boundaries of Autonomy

Prominent defense industry figures, like Palantir co-founder and Anduril investor Joe Lonsdale, have raised concerns about rigid AI policy frameworks, that this question is being framed as a simple yes-or-no decision. At a Hudson Institute event, Lonsdale urged for a more flexible approach to AI in weapons, especially given the rising pressure to keep pace with adversaries like China and Russia, who may be more willing to embrace autonomous technologies. "Before policymakers put these rules in place and decide where the dials need to be set in what circumstance, they need to learn the game and learn what the bad guys might be doing, and what’s necessary to win with American lives on the line,” he argued.

A Shifting Battlefield and Growing AI Integration

The ongoing conflict in Ukraine has highlighted the increasing role of AI in warfare. Ukrainian officials, eager to gain a technological edge, have called for more automation in weapons systems. As defense tech companies test AI in real-world combat scenarios, the lines between human and machine decision-making grow increasingly blurred. With adversaries like Russia keeping their stance on AI arms ambiguous, the global race to develop autonomous weapons shows no signs of slowing down.

What This Could Mean for the World If AI Can Kill

The prospect of fully autonomous weapons raises serious ethical and security concerns for the global community. If AI is granted the authority to make life-or-death decisions, it would represent a violation of Asimov's first law of robotics—never to harm a human being. The lack of clear international regulation on autonomous weapons could lead to a dangerous arms race, where nations feel compelled to adopt lethal AI technology to remain competitive. Without stringent controls, the use of autonomous weapons could destabilize warfare, diminish human accountability, and create a world where life-and-death decisions are no longer in human hands, but in algorithms.