TSA’s facial recognition tech is highly accurate, review says
DHS analysis of its face capture and facial recognition tools summarized testing it conducted of the artificial intelligence-enhanced capabilities used by the Transportation Security Administration, as well as U.S. Customs and Border Protection and Homeland Security Investigations.
The report was conducted after DHS released a directive in September 2023 which required, in part, that all of the agency’s face recognition and face capture technologies be tested to ensure they comply with federal standards and do not produce biased or inaccurate results.
TSA has deployed identity verification tools at more than 80 airports thus far, with the ultimate goal of expanding the technologies to over 400 airports in the coming years. These new units, known at CAT-2 systems, take real-time pictures of travelers — the “face capture” aspect of the tools — and then compare the images against their scanned photo identifications — in this case, in a process known as “one-to-one” facial recognition.
Although travelers can opt out of the process — and TSA places signage in airports disclosing the use of the technologies and outlining that it is optional — lawmakers and privacy rights advocates have expressed alarm about the creeping use of government surveillance tools and their impact on Americans’ civil liberties.
Researchers have also found that facial recognition tools can produce biased results, with Black individuals misidentified at much higher rates than other ethnicities.
In announcing the release of its report, DHS said its analysis “presents more information than ever previously shared about how we use and govern these technologies.”
The DHS Science and Technology Directorate tested TSA’s biometric identification technologies in both 2023 and 2024 using “a very large scenario test with more than 1,600 volunteers.”
The evaluation found that both the face capture and facial recognition aspects of the CAT-2 systems worked more than 99% of the time, and that the results did not significantly vary based on age, gender, race or skin tone.
“On average, the TSA CAT-2 identity verification process took 23 seconds per person,” the analysis said. “It took well under 30 seconds for all demographic groups, and all demographic groups were within a few seconds of the average.”
Self-identified Black or African American volunteers in the testing process had the lowest face matching success rate, with an overall accuracy of 98%.
TSA and DHS officials also told Nextgov/FCW last January that they also conduct limited field testing of the CAT-2 systems, with DHS S&T reviewing collected data for independent analysis of match performance.
Both agencies stressed that the systems do not automatically store data, photos are deleted after a match has been made and any matching issues are remediated by TSA officers. They said, however, that TSA has the ability to change the system so that it can log data for testing purposes.
Democrats and Republicans in Congress have also expressed concerns about TSA’s use of the tools, particularly when it comes to the agency handling travelers’ personal data.
Earlier this month, Republican leaders of the House Homeland Security Committee sent a letter asking the Government Accountability Office to review TSA’s adoption “of biometric identification and AI systems.” That letter came after a bipartisan group of senators similarly asked the DHS Inspector General last November to look into TSA’s authority to operate facial recognition tools, as well as the agency’s handling of travelers’ biometric data.
The latest report also looked at TSA’s PreCheck’s Touchless Identity System — which is still in the pilot phase — and found that its face capture tool only worked 93% of the time due to “an issue with the face detection algorithm, which automatically confirms if a photo actually contains a face the photo is processed for face recognition.” DHS said, however, that the system’s facial recognition tools were accurate more than 99% of the time.
In addition to reviewing TSA’s use of facial recognition, the report analyzed CBP’s use of biometrics for scanning those entering and leaving the U.S., as well as Homeland Security Investigations’ use of the tools to aid in child sexual abuse investigations.
In a LinkedIn post on Jan. 17, former DHS Chief Information Officer and Chief Artificial Intelligence Officer Eric Hysen said all of the department’s analyzed biometric tools included in the report “performed extremely well for diverse demographic groups.”
“For fully operational systems, like ID checks for travelers at airports and ports of entry, the technology worked more than 99% of the time,” he added. “And when minor issues were identified, we acted swiftly to address them.”
]]>