WASHINGTON, D.C. (National Defense Magazine) — The Defense Counterintelligence and Security Agency has proposed using artificial intelligence in its security clearance application process, but a New York-based law firm says questions remain about the technology’s potential benefits.
The agency has led the government’s background check process since 2019, and it is now exploring how AI could speed up the vetting process and reduce applicant backlogs, potentially cutting timelines from months to hours, according to the law firm Tully Rinckey PLLC.
Despite the appeal of faster operations, bringing AI into the government’s “most sensitive” vetting process could be a turning point for both the public and private sectors, it said.
The firm helps candidates applying for security clearances, assisting when clearance is in jeopardy over alleged violations of adjudicative guidelines, said Anthony Kuhn, managing partner at Tully Rinckey.
Adjudicative guidelines are standardized rules used by the government to determine if an individual is trustworthy enough to hold a security clearance. The firm uses 13 of these guidelines to help individuals either obtain or keep their security clearance, Kuhn said in an interview.
The volume of security clearances is cyclical and depends on whether the federal government is hiring. There is an increase in applications when there is a changeover in presidential administrations. However, at any given time, tens of thousands of applicants are stuck in backlogs, Kuhn said.
Artificial intelligence could flag responses during the application process, but there are mitigation factors that could be in play when that occurs, he said. For example, a candidate may say they used marijuana but only for medicinal purposes or because they lived where it was legal. AI is “not going to be able to catch mitigation or understand mitigation,” he said.
“Artificial intelligence may help accelerate parts of the background investigation process, particularly by identifying objective issues that warrant further review,” Kuhn said. “However, AI is only as reliable as the data behind it, and accuracy concerns remain. Security clearance decisions often require human judgment to evaluate context, mitigation and credibility in ways that a machine cannot.”
The very reason the agency wants to incorporate AI — to speed up the process — might have the opposite effect if an AI system creates inaccurate results, Kuhn said.
If the agency allows AI enough authority to flag cases and move the applications to a notice of intent to deny or revoke, then the cases Tully Rinckey responds to and potentially litigates will significantly increase, he said. On the other hand, a human operator can filter those cases out early rather than moving them on to a notice of intent to deny or revoke if there are exceptions to a flagged answer.
“The only positive effect is when you get a very clean application,” such as an individual in the military who has never done drugs, has never been arrested and “didn’t have to answer ‘yes’ to any of the negative questions during the background investigation,” Kuhn said.
Additionally, AI cannot follow up with references, such as an applicant’s prior employer, to confirm the answers on the application, he added.
Like anything in the cyber domain, the highly sensitive vetting process is also susceptible to hackers who could leak applicants’ information, but it depends on the company and how they contract with AI, Kuhn said, adding that it will be interesting to see how preventative steps are enacted to restrict vulnerabilities.
Additionally, new government regulations would be needed to monitor the use and security of AI tools in the vetting process, he added.
“A misconception that people have is they think that AI is safe and they are plugging their information into an AI software platform and getting a response based on their prompts,” Kuhn said. “What they are actually doing in a lot of cases is publishing what would otherwise be confidential or protected information.”
AI is not the only option to help speed up the security clearance process and cut down on backlogs, Kuhn said. More adjudicators could be hired to make cases easier — like removing questions that provoke responses that tend to be reversed — and a more consistent approach could be implemented across the board, he said.
“There is a reason we have not implemented AI yet,” Kuhn said. “It has been implemented in just about everything else.”
