OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy; Human-Machine Interfaces
OBJECTIVE: Develop an artificial intelligence/machine learning (AI/ML) tool to augment declassification reviewers analyzing historical documents for classified equities to more efficiently and accurately determine the final disposition (exclude, exempt, refer, or declassify) of the record. The tool should identify the section(s) of the document more likely to contain classified equities, in accordance with Executive Order (EO) 13526.
DESCRIPTION: Declassification reviewers analyze historical records, in accordance with EO 13526, to prevent the automatic declassification of classified holdings. Due to the amount of material required for review each calendar year, timely reviews are essential to stay on target each month. This ensures no material is automatically declassified at the end of the calendar year prior to review. The government currently employs an enhanced manual review process with digitized paper holdings and keyword search tools to support all levels of declassification review.
PHASE I: Determine feasibility of applying AI/ML algorithms and tools to identify data to remain classified during declassification reviews by comparing to source documents with correct dispositions from archives, security classification guides (SCGs), and declassification guides (DGs). Investigate innovative solutions that can be implemented to support declassification reviewers. Recommend a path forward with AI/ML tools.
PHASE II: Design and develop innovative solutions, methods, algorithms, and concepts to implement AI/ML to support reviewers performing declassification reviews on classified historical records. Create AI/ML algorithms to better predict and isolate pages (or sections) in digitized documents more likely to contain exempted information. The solutions should capture the key areas for new development, suggest appropriate methods and technologies to minimize the time intensive processes, and incorporate new technologies researched. Complete a detailed prototype. Design, build, and test prototype AI/ML tool that would be able to identify areas of text likely to remain classified within a document. Coordinate with the government during prototype design and development to ensure the delivered tool would improve the reviewers’ ability to find exempted material more efficiently and to provide accurate final dispositions. Finally, the solutions should provide a robust demonstration of the tool using historical files found in live production.
PHASE III DUAL USE APPLICATIONS: Scale up the capability from the prototype utilizing the new technologies developed in Phase II into a mature, full-scale, stand-alone capability and integrate across the government workforce. Determine the feasibility of using the AI/ML tool(s) and methodologies with mandatory declassification reviews (MDRs), Freedom of Information Act (FOIA) requests, classification challenges, and/or derivative classification.
- National Security Commission on Artificial Intelligence, 2021. https://www.nscai.gov/2021-final-report
- Executive Order 13526, “Classified National Security Information,” 29 December 2009 https://obamawhitehouse.archives.gov/the-press-office/executive-order-classified-national-security-information
KEYWORDS: Declassification; Artificial Intelligence; Machine Learning; AI; ML; Automatic Declassification Review