Audit Flags Transparency Issues, Bureaucratic Delays in FBI’s Facial Recognition and AI Adoption

An audit recently released on the Drug Enforcement Administration (DEA) and the Federal Bureau of Investigation (FBI) has identified significant challenges in the agencies’ adoption of artificial intelligence (AI) technologies, including facial recognition systems. Conducted as mandated by the Fiscal Year 2023 National Defense Authorization Act (FY 2023 NDAA), the audit highlights critical transparency issues with AI vendors and the potential for bureaucratic delays that could hinder the FBI’s ability to deploy new AI capabilities.

Vendor Transparency Hurdles

One of the standout issues flagged in the audit is the lack of transparency from vendors supplying AI tools, particularly those with embedded facial recognition capabilities. The FBI noted difficulties in obtaining a detailed Software Bill of Materials (SBOM) — an inventory of the components within a software package. This lack of transparency means the agency cannot always verify the presence or behavior of AI functions embedded in commercial products.

Without access to these technical details, the FBI risks unknowingly deploying AI technologies that haven’t been subjected to the agency’s rigorous ethical review process. The audit underscores this as a significant challenge for ensuring the accuracy, fairness, and security of AI applications used in national security operations.

In some cases, the FBI has sought alternative methods to verify AI tools, such as partnering with the National Institute of Standards and Technology (NIST) to test the accuracy of facial recognition systems. However, the audit suggests that more comprehensive measures are necessary to address the transparency gap.

Bureaucratic Delays from Ethical Review Processes

The audit also points to potential delays caused by the evolving regulatory environment and the FBI’s internal AI oversight mechanisms. The FBI’s AI Ethics Council (AIEC) — established to ensure compliance with ethical principles for AI use — paused its review process in 2024 due to new guidance stemming from Executive Order 14110 and anticipated National Security Council directives.

The pause has resulted in a growing backlog of AI use cases awaiting approval. As of July 2024, the average waiting time for AI use case approval had stretched to 173 days, with some cases pending for over 300 days. This represents a stark contrast to the previous average approval time of 40 days before the pause.

The delays could impact the FBI’s ability to integrate new facial recognition technologies and other AI systems, potentially slowing critical national security operations that rely on these tools.

Policy Challenges, Changing Regulations and Other Takeways

The audit highlights the broader challenge of navigating a rapidly shifting policy landscape. New regulations and guidance, such as the upcoming National Security Memorandum on AI, are expected to significantly shape the FBI’s AI governance framework. The constant evolution of these policies poses a challenge for agencies trying to maintain compliance while also accelerating AI adoption.

In addition to transparency and bureaucratic issues, the audit identified several other challenges and progress points in the DEA’s and FBI’s AI integration efforts:

• Funding Constraints: Both agencies cited limited budgets as a major barrier to adopting AI technologies, particularly for research and development.
• Workforce Challenges: Difficulties in recruiting and retaining technical talent, exacerbated by lower government pay scales compared to the private sector, are impeding AI adoption.
• IT Infrastructure: Outdated data architecture and legacy IT systems hinder the seamless integration of AI tools, including those for facial recognition and intelligence analysis.

Despite these hurdles, the audit noted that the FBI has made progress in establishing governance frameworks for AI, such as implementing an AI Policy Directive and maintaining an inventory of AI use cases. The DEA’s Office of National Security Intelligence (ONSI) has taken a more cautious approach, relying on AI tools developed by other intelligence community partners.

Source: DoJ

–

December 20, 2024 – by Cass Kennedy and Alex Perala