Former head of ‘Pentagon’s think tank’ joins Anthropic

The United States has “a tight time window to adapt” to the “civilizational” challenge of adapting to AI, the former head of the Defense Department’s Office of Net Assessment, often referred to as the “Pentagon’s Think Tank,” told Defense One. He’s joining AI company Anthropic as a “strategist-in-residence” to lead analysis of how AI is affecting U.S. institutions and competition with China, the company announced Friday.

James Baker led ONA from 2015 to 2025, when it was temporarily closed by the Trump administration. As the director he advised defense secretaries and national security advisors on the long-term effects of emerging technology on national security, and prior to that role served on the Joint Staff and in other advisory roles.

For decades, ONA played an instrumental role in helping the U.S. military evolve to match broader social, economic, environmental and technological trends. 

Andrew Marshall, a policy strategist in the Nixon Administration, established the Office of Net Assessment in 1973 to take a data-driven, “system-of-systems” approach to understanding future trends, which looked across areas of human activity to better forecast how they might interrelate, such as technology development on military affairs or labor. The office forecast how information technology would greatly increase the speed of warfare and the availability and precision of new weapons, including cyber and electromagnetic effects. These ideas prompted large-scale rethinking of force structure and underscored the need to accelerate acquisition reform across the military.

In its last decade, ONA devoted an increasing amount of time to the implications of accelerating artificial intelligence, especially in the context of Cold War institutions that Congress has been slow to adapt. A 2016 summary study, which formed the basis for an unclassified 2017 Belfer Center examination, saw a “Cambrian explosion” in robotics and artificial intelligence that would make warfare cheaper and faster, and reduce the advantage of expensive investments in so-called “exquisite platforms” like $90 million dollar jets.

That trend is playing out today in Ukraine, where the smaller force is using drones to decimate expensive Russian naval and air defense assets.

But Baker said the national security effects of AI stretch far beyond the military. Only by appreciating the vulnerability of all institutions, including the Defense Department, will society be able to adapt to the changes that are coming.

“We aren’t spending enough time thinking about the implications of recursive self-improvement,” he said, meaning intelligent systems that improve themselves far faster than their creators anticipate. “The greatest risk is the long-term viability of present institutions in war and in peace. That’s one of the questions I came to Anthropic to work on. It’s a multi-decade structural—even civilizational—problem.”

The Defense Department shuttered the ONA office last March, claiming to refocus its personnel on what a spokesperson then called another step in a series of cuts to Defense Department basic research beyond applications for specific weapons and technologies. A Pentagon spokesperson said at the time that the reorganization aimed to allow the Defense Department to better address “pressing national security challenges,” but offered no further explanation. In October, the department reinstated a downgraded version of ONA.

This March, the White House designated Anthropic a supply-chain risk, due to a disagreement between current Defense Department leadership and the company over safe deployment of Anthropic models.

In April, Anthropic announced that it would limit the release of a new AI tool dubbed Mythos to a handful of federal agencies and corporate partners, in order to help the company discover vulnerabilities they might otherwise have overlooked. The number of new vulnerabilities logged in the National Vulnerability Database nearly doubled this month.