AI Warfare: The Future Is Now.

Anthropic's ongoing dispute with the Pentagon not only underscores the inherent dangers of autonomous warfare but also starkly reveals its alarming proximity to realization.

The Convention on Certain Conventional Weapons (CCW), an international forum dedicated to lethal autonomous systems, convenes bi-annually at the United Nations in Geneva. When Branka Marijan attended in November 2017, she anticipated the five-day sessions—typically focused on theoretical discussions of a future dominated by "killer robots"—would follow established patterns. After all, many considered such technology either improbable to develop or unlikely to be deployed. However, Marijan quickly discerned that this particular year marked a significant departure; that once-distant, imagined future had suddenly become more immediate and tangible than ever before.

On the opening day, some delegates viewed a short film titled "Slaughterbots," produced by the Future of Life Institute. The video depicted a fictional defense contractor presenting an AI-powered drone capable of executing unassisted, precise strikes. Its CEO provocatively declared to the audience, “They used to say guns don’t kill people, people do. But people don’t. They get emotional, disobey orders, aim high. Let’s watch the weapons make the decisions.” Marijan recalls a palpable shift in the room's atmosphere, which turned apprehensive. The most unsettling revelation wasn't merely the film's premise, but the stark reality that the Pentagon was already actively developing a similar version of this technology.

This pivotal meeting occurred shortly after the launch of Project Maven, a U.S. Department of Defense initiative leveraging AI for drone surveillance footage analysis. By late 2017, Maven had secured a major tech partner: Google. Marijan, a senior researcher at Project Ploughshares, an independent peace research institute, emphasized, “The systems we were talking about were not futuristic. They were existing platforms that had degrees of autonomy in them, or the capability to select and engage targets based on sensor data and sensor input.”

While the world was already familiar with drone warfare—lethal machines controlled by humans—it was now confronting a future where human involvement might be entirely eliminated. Marijan clarified, “These were not these Terminator-like figures that we were concerned about, but really what was happening with the enablement of autonomy.”

For decades, the U.S. military has consistently championed AI development, a commitment that has, in turn, profoundly reshaped the landscape of warfare.

Nearly a decade subsequent to these revelations, fully autonomous lethal weapons have yet to be deployed by militaries. Nevertheless, these advanced systems are central to a recent high-stakes dispute between the U.S. government and the AI startup Anthropic. Anthropic is striving to uphold two critical "red lines": prohibitions against domestic mass surveillance and against weapons capable of identifying, tracking, and eliminating targets without any human intervention. Since the beginning of the year, Anthropic has distinguished itself as the sole military AI contractor to impose meaningful restrictions on what experts consider one of the ultimate frontiers of AI warfare.

However, amidst evolving alliances, legal battles, and dramatic turns, it's easy to overlook the broader context: AI has long been, and remains, deeply integrated into military operations. Seventy years ago, a pivotal summer meeting among scientists in New Hampshire first alerted the Department of Defense to AI's significant potential for warfare. Since then, AI's influence has expanded exponentially with each passing decade, enabling more rapid and extensive casualties in recent years than ever before.

Even Anthropic itself appears to harbor doubts about the long-term viability of its "red lines," recognizing that historical precedents often suggest otherwise.

The U.S. military's sustained backing of AI development has fundamentally transformed modern warfare. By the 2000s, the technology had advanced to process unprecedented volumes of global data, ushering in a surveillance revolution. The late 2010s further witnessed the emergence of sophisticated machine vision systems, including advanced facial recognition capabilities.

The confrontation between Anthropic and the Pentagon has brought crucial attention to the escalating power of these systems. The dispute originated in January 2026, when Defense Secretary Pete Hegseth mandated a renegotiation of the DOD’s existing AI contracts. The proposed terms eliminated any ambiguities or prior agreements, granting the Pentagon license to utilize companies' technology within the broad and ill-defined scope of "any lawful use." Anthropic, at that time the only AI company authorized to deploy its technology on the Pentagon’s classified networks, voiced its objection.

For a contractor like Anthropic to impose limits on the specific applications of its technology is an uncommon occurrence. Andrew Reddie, an associate research professor of public policy at the University of California, Berkeley, noted, “It’s not government-created technology in the way that the Manhattan Project was,” nor is Anthropic a conventional military supplier akin to Northrop Grumman. He added, “This is one of the pain points that’s made clear when you’ve got this startup ecosystem engaging with the Pentagon directly.” Reddie also highlighted that even within Silicon Valley, there is "a lot of disagreement" regarding where such limits should be drawn.

The ultimate victor in this confrontation remains uncertain. As a negotiation tactic, the DOD labeled Anthropic a military supply chain risk in March, prompting President Donald Trump to declare a ban on all government agencies from utilizing its Claude system. While the relationship has reportedly softened somewhat since then, particularly with the release of Anthropic’s cybersecurity-focused model, Mythos, a legal battle is still unfolding. Anthropic declined to comment for this article.

Regardless of the outcome, this debate has elevated the concept of "fully autonomous weapons" into public discourse in a novel way. Yet, AI's pervasive influence in military operations has shown no signs of abating over recent decades.

“We’ve kind of crossed the rubicon while we pretend that we haven’t,” Reddie asserted.

Central to these discussions is DOD Directive 3000.09, one of the few policies governing the deployment of lethal autonomous weapons. Originally drafted in 2012, it defines such a system as one that, “once activated, can select and engage targets without further intervention by an operator.” The directive mandates that both fully autonomous and semi-autonomous weapons must be designed to enable humans to “exercise appropriate levels” of judgment regarding the use of force.

Hamza Chaudhry, who leads AI and national security at the Future of Life Institute, stated that the directive established the “first policy on the use of autonomy in warfare.”

“Even without full autonomy, AI compresses kill chains to mere seconds.”

However, depending on the interpretation of this definition, certain missile defense programs might have arguably crossed that threshold decades ago. Consider the Phalanx CIWS, an automated weapon system resembling a large gun, designed to protect naval vessels from incoming missile attacks. Such a system necessitates millisecond response times, rendering human intervention impractical.

The distinction, some experts argue, lies in these systems operating exclusively within a defensive, fixed environment. They are engaging, according to this interpretation, but not making independent decisions—merely reacting to an immediate threat. Reddie elaborated, “The ‘and’ is doing a lot of work inside of that statute — we have systems that can decide and systems that can engage but you can’t have a system that does both.”

Maddy Batt, a legal fellow at Tech Justice Law, warned, “Even without full autonomy, AI compresses kill chains to mere seconds so that humans are not actually making the assessments that international humanitarian law requires to prevent civilian harm. When humans’ failure to do that results in civilian death, that is a war crime.”

The demarcation between offensive and defensive capabilities also remains ambiguous. Reddie recounted, “One of my favorite exercises with my students is putting up a military technology and then asking, ‘Is this defensive or offensive?’” He cited the example of a nuclear weapon in a silo, which some consider defensive due to its deterrent purpose, while others classify it as offensive given its design to strike foreign targets. “Just because its primary function … is defensive in nature doesn’t mean the technology itself is defensive in nature,” he concluded.

Indeed, certain missile defense systems might have crossed the line of autonomous response decades ago.

Sorin Adam Matei, a professor at Purdue University, offered a more blunt assessment: “You cannot fight a war only in defense.”

In 2023, the government’s Chief Digital and Artificial Intelligence Office (CDAO)—the central hub for all DOD AI operations—issued an update to DOD Directive 3000.09. However, this revision failed to resolve the document's fundamental ambiguities. The Biden administration subsequently published a memorandum on AI and national security in 2024, outlining rules for AI's use in specific national security contexts, a policy that remains in effect even under the Trump administration. Despite this, the Pentagon has experienced significant upheaval. The CDAO is currently undergoing a substantial restructuring, leading to its increased isolation within the DOD, and the office now reports to Emil Michael, who serves concurrently as the DOD’s undersecretary of research and engineering and the department’s CTO.

International initiatives, such as the Convention on Certain Conventional Weapons and the Campaign to Stop Killer Robots, have similarly struggled to achieve substantial progress. Marijan informed The Verge that while the CCW has been instrumental in helping smaller nations grasp the complexities of AI warfare, overall progress has been “very slow and we haven’t seen concrete agreement, particularly among the major countries and the more sophisticated militaries.” Despite interest from some nations in a ban on lethal autonomous weapons, an official international definition for the term remains elusive. Sarah Shoker, a senior research scholar at the University of California, Berkeley, and former lead of OpenAI’s geopolitics team, explained that this situation “often results in a lot of people talking past each other … and some countries find the lack of a binding instrument to be to their advantage.”

While some countries have expressed interest in banning lethal autonomous weapons, a universally recognized international definition for the term has yet to be established.

Shoker further observed, “I think most people — policymakers, civil society members … who attend these meetings are likely tired. It’s been over a decade, and there is really no agreement.”

Regardless of the state of autonomous weapons prior to 2017, Project Maven dramatically accelerated their timeline and fundamentally altered the terms of the debate. The program marked the first undeniable involvement of the ostensibly "don't be evil" Silicon Valley of the 2010s in warfare, prompting strong reactions from Google employees and the wider public. Matei characterized it as “the first national conversation we had about the role of AI technologies in military operations.”

Project Maven originated with a memo from Robert Work, then deputy defense secretary, who established an “Algorithmic Warfare Cross-Functional Team.” This team was tasked with training AI algorithms for deployment to war zones, potentially within the same year. Google was subsequently awarded the contract.

Internal resistance within Google quickly mounted. In April 2018, approximately 4,000 employees signed a petition demanding that the company withdraw from the “business of war.” The petition referenced assurances