OpenAI’s partnership with Anduril puts its technology into a setting the company had publicly kept at a distance: battlefield defense. The program is focused on helping US and allied forces respond to drone attacks, but it also shows how quickly OpenAI’s position on military work has changed.
A fast change in policy
At the start of 2024, OpenAI’s rules left little room for military applications. The company prohibited use of its models for “weapons development” or “military and warfare.”
That stance shifted on January 10, when The Intercept reported that OpenAI had softened those restrictions. The newer language barred using the technology to “harm yourself or others” through developing or using weapons, injuring others, or destroying property.
OpenAI then said it would work with the Pentagon on cybersecurity software, while maintaining that it would not work on weapons. In October, the company went further in a blog post about national security, arguing that AI, used properly, could “help protect people, deter adversaries, and even prevent future conflict.”
The Anduril agreement takes that evolution into a more concrete arena. OpenAI says its technology will be used directly in battlefield-related systems, even as the company frames the work as defensive and consistent with its policies.
What the Anduril partnership covers
Anduril makes AI-powered drones, radar systems, and missiles. Under the new partnership, OpenAI will support systems meant to help US and allied forces defend against drone attacks.
OpenAI’s role, according to the announcement, is to help build AI models that “rapidly synthesize time-sensitive data, reduce the burden on human operators, and improve situational awareness.” The stated goal is to help take down enemy drones.
Specific details of the program have not been released. Liz Bourgeois, an OpenAI spokesperson, said the work will be narrowly focused on defending US personnel and facilities from unmanned aerial threats.
“This partnership is consistent with our policies and does not involve leveraging our technology to develop systems designed to harm others,” Bourgeois said.
An Anduril spokesperson did not give specifics about the bases around the world where the technology will be deployed. The spokesperson said the system will help spot and track drones and reduce the time service members spend on dull tasks.
Why the shift matters
OpenAI has long described itself as an organization built around the mission of ensuring that artificial general intelligence benefits all of humanity. Its founding charter placed that mission at the center of the company’s identity.
That history makes the defense move different from similar choices by companies that were built around military technology from the start. Defense-tech firms can present warfare work as their core business. OpenAI is moving from a public posture that once treated military work as incompatible with its mission to a position that says some national security work can serve that mission.
In its October 24 blog post, titled “OpenAI’s approach to AI and national security,” the company tried to explain that change. The post was published the same day the White House issued its National Security Memorandum on AI, which ordered the Pentagon and other agencies to increase their use of AI, partly to counter competition from China.
OpenAI wrote: “We believe a democratic vision for AI is essential to unlocking its full potential and ensuring its benefits are broadly shared.” It added: “We believe democracies should continue to take the lead in AI development, guided by values like freedom, fairness, and respect for human rights.”
The company listed possible national security uses such as efforts to “streamline translation and summarization tasks, and study and mitigate civilian harm.” It also said its technology remained barred from being used to “harm people, destroy property, or develop weapons.”
The defense-tech backdrop
OpenAI’s move is not happening in isolation. The broader technology sector has been moving toward defense work for years.
Venture capital firms more than doubled their investment in defense tech in 2021, to $40 billion. Companies such as Anduril and Palantir showed that the Pentagon would pay for new technologies, though the path involved persuasion and litigation.
Employee resistance to warfare work has also changed. Opposition was highly visible during walkouts at Google in 2018. Some of that resistance softened after Russia invaded Ukraine in 2022, according to several executives in defense tech who said the “unambiguity” of that war helped them attract investment and talent.
Large technology companies have also competed for Pentagon cloud computing contracts. Amazon, Google, and Microsoft, which is OpenAI’s partner and investor, have all pursued that market. The source article notes that such relationships no longer draw the same level of outrage and scrutiny they once did.
OpenAI’s own business context is part of the picture. The company expects $5 billion in losses and is reportedly exploring new revenue streams, including advertising. Defense contracts may therefore represent another possible market, although OpenAI’s technology is not simply cloud storage.
Questions that remain unresolved
The central tension is whether a defensive battlefield system can be cleanly separated from weapons work. OpenAI says the Anduril program is about protecting personnel and facilities from unmanned aerial threats. Critics argue the boundary is less clear.
Heidy Khlaaf, a chief AI scientist at the AI Now Institute and a safety researcher who authored a paper with OpenAI in 2022 about possible hazards of its technology in contexts including the military, said the newer policies stress “flexibility and compliance with the law.” In her view, the pivot “ultimately signals an acceptability in carrying out activities related to military and warfare as the Pentagon and US military see fit.”
Khlaaf also challenged the distinction between offensive and defensive systems. “Defensive weapons are still indeed weapons,” she said. They “can often be positioned offensively subject to the locale and aim of a mission.”
That concern points to the larger issue raised by the Anduril deal. OpenAI can state its intended use and write limits into its policies, but defense customers operate under different pressures and rules. When the customer is the US military, technology companies may have limited control over how their products are ultimately used.
The partnership therefore marks more than a new contract. It is a test of how OpenAI defines responsible AI when its models are used to process battlefield data, identify threats, and make military decision-making faster.