Over the past year, the PIDB has had an ongoing opportunity to evaluate tools for automating classification and declassification at a number of federal agencies, including the Department of Defense, the State Department, and the CIA. We have also taken time to speak with experts from industry in order to understand the broader emerging landscape of tools for Artificial Intelligence-driven and machine-assisted information management.
One thing is clear – the capacity to radically rethink and reform how we manage classified information is upon us. A technological transformation long envisioned (and long advocated for) by the PIDB is no longer conjecture. It is now well within the realm of the possible — and for some parts of government, these capabilities have already been put into practice.
At the same time, we have learned that even in places with the most expansive resources, and even when information management programs have the best benefit of emerging technology, the government is still thinking too small. It continues to employ relatively rudimentary, highly fragmented systems that do not appropriately encompass the full lifecycle of information.
Managing classified information is not a single process. There are many facets to the existing workflow for classification and declassification that require specialized applications with different customers and capabilities. A more holistic, technology-driven approach will need to account for these various business processes that agencies seek to modernize and automate.
When it comes to current authorities for classification and declassification, defining this “technology stack” is a notable vacuum in the system. The current Executive Order on Classified National Security Information, which dates back to 2009, lays some of this burden on the National Declassification Center (NDC) at the National Archives. The then newly established NDC was given the assignment to bring technology to bear on declassification, a duty it has never been positioned, funded, or staffed to perform.
At that time, transparency advocate, Steven Aftergood wrote that the NDC, as conceived, already, “seems like an Industrial Age solution to an Information Age problem.” The passage of time has only shown this to be a more accurate assessment, as we find ourselves in 2025 with the core technology problems for classification and declassification still broadly unaddressed.
The limitations of current technological solutions have become evident following President Trump’s Executive Orders directing the declassification of records related to the assassinations of President John F. Kennedy, Senator Robert F. Kennedy, and Dr. Martin Luther King Jr. While the Board commends the release of these historically significant records—despite the unacceptable delays that preceded their disclosure—it envisions a framework in which public transparency is achieved without reliance on extraordinary presidential intervention.
To this end, the government must establish a centralized and automated system capable of autonomously reviewing classified records and preparing them for public release upon the expiration of their classification period. Such a system would ensure a continuous and predictable flow of declassified historical information for public access. In the absence of this reform, delays will persist, necessitating ad hoc presidential action, as is likely to be the case with the Congressionally mandated declassification review of records related to Unidentified Anomalous Phenomena and with a substantial body of records concerning the September 11 terrorist attacks that are approaching their 25-year mandatory declassification deadline. The PIDB is committed to drawing attention to these systemic deficiencies, with particular emphasis on the challenges surrounding the timely release of September 11 records.
Unless addressed, this technology gap – and the lack of a central, directive authority to drive agencies towards more holistic solutions — will continue to haunt future declassification efforts, both those undertaken in the broad public interest, as well as the daily work that must continue to ensure that Americans have appropriate access to information about the activities of their government.
9/4/25, ALL Government systems are antiquated!! I was at the Social Security office for the last two days and could not log in and wondering if there is an issue? So, lo and behold a white bus with nothing on the bus, to determine who they were? Out, come, a lot of illegals to get social security cards ???
I went up to the window and asked why I wasn’t called, after been waiting 4 hours??? ????
This is an important post – and one I hope agencies read and take note. Classification and Declassification programs are still far too agency-centric, with officials arduously guarding ‘their’ information. This means that information remains stovepiped. We’ve seen the first generation of reform with NGA’s ground-breaking effort to consolidate all its classification guides into a single guide – one that has mechanisms for regular updates and metadata connection to products that its staff create. But the next step is overdue. Today, departments are all going solo in developing AI/ML classification and declassification guides. Neither will help inter-agency access and use of information. Over-classification, under-classification, mis-classification result. And the problems are worse when it comes to declassification.
Solving the problem with the classification system and massive amounts of classified electronic records requires and end-to-end solution. The problem can’t be solved with the automation of classification alone.
Automation of classification must include automation of the classification process, the continued management of classification guides where all guides are in a central repository and continually updated. Classification metadata created at the time of document classification is carried with the document so as guides are revised (which must be a mandated periodic process), information no longer currently classified is automatically declassified so documents can be downgraded as soon as the most sensitive information is reduced in classification and declassified with the information classified at the time of origin is no longer classified.
For the system to be autonomous, the guidance MUST be continually managed and the document metadata used to manage the classification of documents created with the system. For documents created prior to the automation of classified document creation, the system must have a means to correct the contextual errors common in AI systems.
To obtain this level of solution appropriate requirements for a system must be created so AI experts can create a system tailored to help resolve the problem of an archaic classification system and overwhelming volume of classified electronic records.