You are here

Mandatory Declassification Review (MDR) Natural Language Processing (NLP) Tool

Description:

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy

 

OBJECTIVE: Finding a solution to assist AFDO’s Mandatory Declassification Review (MDR) program in increasing efficiencies, achieving consistency of line-by-line review and redaction of information to remain classified, and promote cost savings through use of new technologies and industry best practices. A successful solution has potential to become a program of record for the program, upon completion of the appropriate acquisition process, obtaining an Authority to Operate (ATO) via Certification and Accreditation (C&A), with an established funding line, and deployment to an approved host location.

 

DESCRIPTION: There are six milestones a selected company would need to achieve and gain approval by AFDO, to meet the aforementioned objective:

Milestone 1: Refine and enhance the AI/ML models for line-by-line reviews for cases assigned under the Mandatory Declassification Review program.

Milestone 2: Conduct extensive testing and evaluation of the solution in collaboration with AFDO (MDR) personnel.

Milestone 3: Optimize the solution based on feedback and lessons learned during testing.                         

Milestone 4: Develop a user-friendly interface and integrate the solution with any potential AFDO workflow system.

Milestone 5: Complete documentation for a deployment plan, addressing security and operational requirements, and any other required documentation.

Milestone 6: Upon approval by AAI Director & AFDO Leadership, ensure compliance requirements are met to deploy tool to specified environment/platform.

 

PHASE I: This is a Direct-to-Phase-II (D2P2) topic, no Phase I awards will be made because of this topic. To qualify for this D2P2 topic, the Government expects the applicantOfferor to demonstrate feasibility by means of a prior “Phase I-type” effort that does not constitute work undertaken as part of a prior SBIR/STTR funding agreement.  ApplicantOfferors are expected to provide a white paper containing the following information on Artificial Intelligence, Machine Learning such as:

1. Refine and enhancing AI/ML models for line-by-line reviews for cases assigned under Mandatory Declassification Review (MDR).

2. MDR Tool Prototype Outline: AI/ML driven line-by-line review capabilities, highlighting areas requiring classification determination based on relevant SCDGs.

3. Optimization of the solution based on feedback and lessons learned during "Phase-I-type".

AFDO strongly encourages companies to submit Direct-to-Phase II proposals to facilitate a demo and hands-on use of a prototype by the end of the phase. Unlike Phase I submissions, Direct-to-Phase II offers extended time and allocated funds, enabling companies to better meet the Government’sour specific prototype requirements.

 

PHASE II: AFDO leadership are looking to establish a tool for this organization’s newly owned requirement of handling the Department of the Air Force (DAF) Mandatory Declassification Review (MDR) program. MDRs may either be requested direct from a public requestor or referred to DAF (AFDO) for review from another government agency, based on potential Air Force equities. AFDO reviewers must conduct a line-by-line review and provide specific alignment to Executive Order 13526 based on the appropriate exemption selected in Security Classification and Declassification Guides (SCDGs).

With this newly acquired program, there is a large backlog of cases for AFDO to review, task out to other organizations/agencies, all while new cases continue to trickle in. Given the typical content of the requests from the public and other government agencies, the sponsoring organization has a need to:

- Respond to requests in a quicker manner;

- Provide consistency & accuracy during the document review;

- Identify similar topics from previous cases that may apply in the future;

- And identify potential equity of other organizations/agencies during intake of the case.

The sponsoring organization’s current process is almost entirely manual, therefore this topic’s focus is to truly enhance the review process altogether, while ensuring compliance with mandated standards. Information requested under MDR may still retain its classification, and therefore AFDO must pay close attention to information released, as the impact of releasing current classified information could cause up to exceptionally grave damage to national security.

 

PHASE III DUAL USE APPLICATIONS: Phase III would incorporate the solution into the daily business processes at AFDO, including:

      -     Transition of AFDO reviewers to new innovative process and addressing any issues;

      -     Workflow incorporation, adding in the administrative piece for MDR’s initially and upon reviewer decision notification;

      -     Rule development (adjustment) and management of the tool in-house;

      -     Training the tool, encompassing continued updates and feeding the tool data sets;

      -     Full deployment to approved DAF host location.

 

REFERENCES:

  1. Atomic Energy Act of 1954, as amended;
  2. 10 CFR Part 1045, Subpart A-D;
  3. 32 Code of Federal Regulations (CFR) Parts 2001 and 2003 Classified National Security Information;
  4. Executive Order 13256;
  5. DoDI 5210.02 Access to and Dissemination of Restricted Data and Formerly Restricted Data;
  6. DoDM 5200.01 DoD Information Security Program;
  7. DAFMAN 16-1404 Information Security Program;
  8. Air Force Declassification Guide for Historical Records;

 

KEYWORDS: Artificial Intelligence; AI; Machine-Learning; ML; Contextual Search; Natural Language Processing; NLP; Mandatory Declassification Review; MDR; Security Classification Guide; SCG; Declassification Guide; Executive Order 13526; Exemption; Restricted Data; RD; Formerly Restricted Data; FRD; Atomic Energy Act; Exclusion; Line-by-line; Redaction; E.O. 13526 Section 1.4; E.O. 13526 Section 3.3 (b); E.O. 13526 Section 3.3 (h);

US Flag An Official Website of the United States Government