A once-narrow defense technology question has become a broader argument about power, accountability, and military speed: should AI weapons ever be allowed to decide to kill without a human making the final call?
The answer coming from Silicon Valley is no longer simple. Some defense tech leaders still frame human accountability as essential. Others argue that a rigid ban could leave the U.S. at a disadvantage if China or Russia moves faster.
The debate is moving beyond a simple yes or no
In late September, Shield AI co-founder Brandon Tseng told TechCrunch that weapons in the U.S. would never be fully autonomous, meaning an AI algorithm would make the final decision to kill someone. His view was blunt: “Congress doesn ’t want that,” and “No one wants that.”
Five days later, Anduril co-founder Palmer Luckey took a more skeptical view of the arguments against autonomous weapons during a talk earlier this month at Pepperdine University. He challenged the moral framing often used by critics, saying opponents rely on phrases that sound persuasive in a short clip.
Luckey’s example was designed to question whether older weapons are necessarily more ethical. “And my point to them is, 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?” he said.
Anduril later sought to narrow that interpretation. Shannon Prior, a spokesperson for Anduril, said Luckey did not mean robots should be programmed to kill people on their own. She said his concern was about “bad people using bad AI.”
Human control is still the public baseline
The more cautious view has deep roots in how Silicon Valley defense companies describe their technology. Trae Stephens, Luckey’s co-founder at Anduril, told Kara Swisher last year that the technologies being built were intended to help humans make the right decisions.
Stephens described a system in which an accountable and responsible party remains involved in lethal decisions. The Anduril spokesperson denied that Luckey and Stephens were in conflict, saying Stephens did not mean a human must always make the call, only that someone must be accountable.
That distinction matters. A weapon can contain more AI without necessarily removing human responsibility. But the harder question is where responsibility sits when software identifies, acquires, and fires on a specific human target without human intervention.
The source article makes clear that the U.S. military currently does not purchase fully autonomous weapons. It also notes that some weapons, including mines and missiles, can operate autonomously in a different sense. That is not the same as an AI-enabled system selecting and attacking a specific human target on its own.
U.S. policy leaves room for ambiguity
The government position is not a clean prohibition. The U.S. does not ban companies from making fully autonomous lethal weapons. It also does not explicitly ban companies from selling such weapons to foreign countries.
Last year, the U.S. released updated guidelines for AI safety in the military. Those guidelines have been endorsed by many U.S. allies and require top military officials to approve any new autonomous weapon. But the guidelines are voluntary.
Anduril said it is committed to following the guidelines. U.S. officials, meanwhile, have repeatedly said it is “not the right time” to consider a binding ban on autonomous weapons.
That leaves the defense technology industry operating in a space shaped by guidance, strategic pressure, and politics rather than a hard legal wall. Companies can argue they are not setting policy while still building systems that may shape what policymakers eventually consider acceptable.
Investors are pressing for a more flexible dial
Last month, Palantir co-founder and Anduril investor Joe Lonsdale also signaled a willingness to consider fully autonomous weapons. At an event hosted by the think tank Hudson Institute, he objected to treating the issue as a binary choice.
Lonsdale described a hypothetical in which China has embraced AI weapons while the U.S. must “press the button every time it fires.” His point was that policymakers may need to think in terms of degrees of autonomy rather than a single yes-or-no rule.
“You very quickly realize, well, my assumptions were wrong if I just put a stupid top-down rule, because I’m a staffer who’s never played this game before,” he said. “I could destroy us in the battle.”
When TechCrunch asked for more comment, Lonsdale said defense tech companies should not be the ones setting the agenda for lethal AI. “The key context to what I was saying is that our companies don’t make the policy, and don’t want to make the policy: It’s the job of elected officials to make the policy,” he said. “But they do need to educate themselves on the nuance to do a good job.”
He also rejected the idea that the policy question is simply “fully autonomous or not.” He argued there is “a sophisticated dial along a few different dimensions” for what a soldier does and what the weapons system does.
War and lobbying are changing the pressure
Activists and human rights groups have long tried and failed to establish international bans on autonomous lethal weapons. The U.S. has resisted signing those bans.
The war in Ukraine may have weakened the activists’ position. According to the source article, the war has produced combat data and a battlefield where defense tech founders can test technology. Companies already integrate AI into weapons systems, although a human still makes the final decision to kill.
Ukrainian officials have also pushed for more automation. Mykhailo Fedorov, Ukraine’s minister of digital transformation, told The New York Times, “We need maximum automation,” adding, “These technologies are fundamental to our victory.”
For many in Silicon Valley and Washington, D.C., the central fear is that China or Russia deploys fully autonomous weapons first. At a UN debate on AI arms last year, a Russian diplomat said, “We understand that for many delegations the priority is human control,” then added, “For the Russian Federation, the priorities are somewhat different.”
That fear is helping move the conversation from principle to policy influence. At the Hudson Institute event, Lonsdale said the tech sector needs to “teach the Navy, teach the DoD, teach Congress” about AI’s potential to “hopefully get us ahead of China.”
Anduril and Palantir are already trying to get Congress to listen. According to OpenSecrets, the two companies have cumulatively spent over $4 million in lobbying this year.
The result is a debate with no settled center. Fully autonomous AI weapons remain outside current U.S. military purchasing, but they are not clearly banned. Silicon Valley’s defense leaders are now arguing over how much autonomy is too much, who should decide, and whether the greater danger is moving too fast or moving too slowly.