Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The writer is a professor and director of Perry World House at the University of Pennsylvania and a former deputy assistant secretary at the US Department of Defense
Last Friday, the Pentagon announced that it was classifying AI lab Anthropic as a supply chain risk after contract negotiations broke down. The label meant that not only would Anthropic lose its government contracts but that it was now grouped with Chinese companies like Huawei as a threat to US national security.
Less than 24 hours later, however, Anthropic’s technology was reported to have been used in Operation Epic Fury against Iran. The dissonance of the tech company being labelled a supply chain risk even as it aided the US military was the latest chapter of an unnecessary, yet arguably inevitable, saga as AI is increasingly incorporated into the military.
Anthropic versus the Pentagon is a dispute about the role of government masquerading as a fight about policy.
There was no prior disagreement between the two about any existing AI use cases by the American military. Nor were there public Pentagon plans to direct Anthropic to work on projects that would cross the red lines the AI company had set on the use of its tools for mass surveillance and autonomous weapon systems.
Moreover, OpenAI signed a deal with the Pentagon on Friday evening that contains 99 per cent of what Anthropic wanted and is working to amend the agreement to further strengthen the provisions that will prevent misuse of its technology.
These facts suggest that the dispute reflects a breakdown in trust. The Pentagon did not trust that Anthropic’s tools would be available when needed for important national security uses; Anthropic did not trust the Pentagon to use its technology responsibly.
This raises questions about the level of control that the private sector can expect to wield when the military acquires its AI tools. The technology is largely regarded as predominantly civilian, unlike weapons of war such as tanks or missiles. But does that mean AI companies are entitled to more say about the way it is used? After all, companies like Lockheed don’t expect to be able to tell the Pentagon where they are allowed to launch missiles or deploy F35s.
There are, however, notable differences between typical military systems, which are essentially mature when purchased, and AI systems, which are constantly evolving. For example, Anthropic is almost certainly correct that its technology is not yet ready for incorporation into autonomous weapon systems.
It stands to reason that Anthropic would want input into the uses of its systems in order to be effective.
However, this is distinct from having the right to make decisions about appropriate use of force. As the Pentagon has correctly argued, there are already detailed policies on autonomous weapon systems. These include DoD Directive 3000.09, which my former office updated in 2023. There are also legal obligations surrounding the use of force that should prevent dangerous battlefield uses of immature technology.
Perhaps this dispute was inevitable. Today’s weapon systems already feature large degrees of autonomy and have for 40 years, including systems with no human supervision after launch. But the prospect of AI deciding when and how to use force presents new challenges for processes designed in the industrial era.
As a general-purpose technology, AI is reshaping economies and societies around the world. Militaries like the US may have policies governing the incorporation of autonomous weapon systems but many have not been put to the test. And the leaders of today’s technology companies are used to being able to name their terms when negotiating.
Regardless, it is disappointing to see the way in which the dispute has escalated. Anthropic was the first frontier AI company performing classified work to support American national security. Other AI companies will take note of the Pentagon’s legally dubious claim that Anthropic is now a supply chain risk in an attempt to salt the earth over a contract dispute. This could make the pathway to some private partnerships harder in the future. That would be a shame. The Pentagon has had a mutually beneficial relationship with American technology companies since the second world war.
If Anthropic and the Pentagon cannot reach a resolution then the real winner will be the countries hoping to topple America’s AI and military supremacy, especially China. The US military will lose access to the talent and technology of one of the world’s top AI companies. Let us hope they can reach a deal where everyone wins.