Privacy and Civil Liberties Oversight Board convenes public forum on AI and counterterrorism; experts urge measurable safeguards

The Privacy and Civil Liberties Oversight Board opened a public forum to examine the role of artificial intelligence in counterterrorism and related national security programs. Chair Sharon Bradford Franklin said the board’s role is to ensure that “as our government incorporates AI tools into its efforts to protect the nation from terrorism it is our role to ensure that those government strategies also protect individual rights and Liberties.”

Panelists and a pre‑recorded statement from Senator Mike Browns emphasized both the promise and risks of AI for intelligence work. Senator Browns urged that “measuring these kinds of performance metrics must be a part of the intelligence committee’s oversight” as agencies adopt tools to process large, multi‑source data sets.

Experts from government and the private sector urged a risk‑based, context‑sensitive oversight approach. Alondra Nelson, a social‑science scholar and former acting director at OSTP, warned that ‘‘we must be strategic in our oversight of the government's use of AI for counterterrorism purposes’’ and highlighted risks including biased training data, unequal facial‑recognition performance across demographic groups, and opaque decisionmaking.

Representatives of technical and standards organizations called for concrete measurement and evaluation. A senior NIST speaker summarized the agency’s work on the AI Risk Management Framework and said, “if we cannot measure it we cannot improve it,” urging development of measurable tests and playbooks for agencies to operationalize safeguards.

Panelists also debated operational tradeoffs. William Usher of the Special Competitive Studies Project and others underscored AI’s potential to speed analysts’ work — for example, triaging thousands of reports or translating and prioritizing captured archive material — while cautioning that agencies must preserve privacy protections and human review for high‑risk decisions. Usher told the board he would “encourage the president Congress and this board not to prematurely tie the ic's hands,” arguing the U.S. must preserve capability while enforcing safeguards.

A recurring subject was the use of large commercial ‘‘foundation’’ models trained on massive datasets. Panelists generally said the intelligence community will likely make use of commercially developed models because building equivalent systems in‑house is costly, but urged agencies to retrain and validate models to meet privacy and accuracy standards, and to adopt clear policies about the handling of U.S.‑person data.

The forum highlighted several concrete red lines and priorities: panelists warned against civilian deployment of real‑time facial recognition without strict limits; opposed reliance on automated systems for lethal targeting without robust commander‑intent and human oversight; and asked the board to prioritize oversight of predictive systems or AI uses that materially change collection, retention or dissemination of U.S.‑person data.

The board and panelists agreed on procedural next steps rather than new policy mandates: sustained measurement (predeployment testing and ongoing evaluation), stronger procurement and acquisition safeguards, continuous privacy impact assessments, and contestability mechanisms to allow affected individuals and oversight bodies to challenge AI‑driven decisions. The forum record will inform the board’s scope of oversight as agencies implement the White House executive order and the forthcoming national security memorandum on AI.