Second Annual Jervis Conference Examines Overclassification and the ‘Dark State’

The impact of classification and government secrecy on American democracy and national security is profound. To discuss the ramifications of these practices, as well as the role played by Artificial Intelligence and intelligence successes and failures, on April 19, SIPA hosted the second annual conference in honor of the late Columbia political scientist Robert Jervis, titled, “The Dark State? Government Secrecy and American Democracy.” The event was presented by the Institute of Global Politics, the Saltzman Institute of War and Peace Studies, the Public Interest Declassification Board, and the Institute for Social and Economic Research and Policy. 

Jervis taught at Columbia for more than 40 years, and left a lasting legacy on the field of international relations and the US intelligence community.

In her opening remarks, SIPA Dean Keren Yarhi-Milo emphasized Jervis’s legacy, including the legion of scholars who followed in his footsteps, seeking to understand how intelligence agencies work or how political leaders and their staffs make decisions during crises. “Without seeing the original documents and analyzing the actual minutes of closed-door meetings, we could not find the patterns that would otherwise stay hidden,” said Yarhi-Milo. “We would lack that contextual analysis and level of detail that case studies in political science require. As someone who works in this area, I’m immensely grateful.”

Image

The conference comprised four panels, the first three of which addressed different facets of the theme of a “Dark State.” The first, moderated by, Carter Burwell, a lawyer and national security specialist who has held multiple positions in the federal government, delved into the legal and political foundations of government transparency and secrecy. These legal experts discussed the tension between the need for government secrecy in national security and the democratic imperative for transparency.

“Unitary executive power actually can undermine the ability of the people to learn about what is going on, including presidential misdeeds,” said Heidi Kitrosser, William W. Gurley Professor of Law at Northwestern, adding that “courts have tended to be extraordinarily deferential, very much afraid to second guess the executive judgment.”

Ben Wizner, who directs the ACLU’s Speech, Privacy, and Tech Project, made the connection between secrecy and keeping the government accountable. He observed that “all too often in national security litigation, disputes over secrecy are really operating as a proxy for disputes over accountability.” 

David Pozen, the Charles Keller Beekman Professor of Law at Columbia Law School, sounded the alarm on Congress’s low appetite to legislate a classification regime. “It's striking that more members of Congress don't even try to invoke their powers under the speech and debate clause to push back against what they see as excessive executive branch secrecy,” he said.

The second panel, moderated by author and journalist Sue Halpern, focused on the use and potential misuse of artificial intelligence in managing sensitive information. Panelists discussed the promise and pitfalls of employing AI in the declassification process.

Hanna Wallach, a partner research manager at Microsoft Research, emphasized the need for government and technology companies to collaborate in developing effective systems, while Camille François, a lecturer at SIPA, noted that developing these AI systems will need to be a deliberate process. “Writing guidelines for humans is extraordinarily different from writing guidelines for large language models,” François said. “It begs the question of what type of guidelines we want to write, and for whom.”

What often worries me about misinformation and disinformation is not just its increased prevalence, but also that human beings increasingly don't care about the truth.

— Michael Collins, Chair of the National Intelligence Council

Carmen Medina, former CIA officer and owner of MedinAnalytics, reiterated the importance of reforming the existing classification systems: “People in the national security community need to ask themselves hard questions about to what extent the value of secrecy outweighs these inane and Byzantine classification systems,” she said.

The third panel, moderated by Columbia political scientist Elizabeth Saunders, addressed the crisis of overclassification and the potential to reform and add greater transparency to the system. Joseph Lambert, a former director for information management services at the CIA, drew upon his experience working in three intelligence agencies to discuss the complexities and misguided assumptions of declassifying intelligence.

Timothy Naftali, a senior research scholar and adjunct professor at SIPA, said the fear of making mistakes makes it “easier to over-classify than to under-classify.” He added that “for the sake of transparency and democratic resilience, it's okay to make a few mistakes with the larger objective of building some trust that our government is incapable of keeping a grand conspiracy secret forever…. The best way to hold the powerful accountable is for them to know they can't get away with something.”

Image

Michael Collins, chair of the National Intelligence Council, stressed the importance of transparency during a time when misinformation is prevalent, “What often worries me about misinformation and disinformation is not just its increased prevalence, but also that human beings increasingly don't care about the truth,” he said. “Then, we're really in an existential problem.” 

The conference concluded with a final panel, “Reflections on What We Have Learned,” with Matthew Connelly, professor of international and global history at Columbia, touching on a few key takeaways and lingering questions. Page Fortna, Columbia political scientist and director of the Saltzman Institute, offered a few provocations to close out the conference: Why is it that the courts in the United States are so reluctant to push back on the executive branch with respect to security and secrecy? How does AI intersect with society and societal priorities? And what can be done on the policy side in terms of reform of over-classification?

“Robert Jervis was deeply committed to the issue of sharing and declassifying information while still sensitive to the need for governments to keep some secrets,” Fortna added. “There's a sort of mutual disdain between the academic and the policy world but that's not been true here at Columbia. And that is largely due to Bob Jervis and example and the culture that he set."

Watch the full program: