Form UnNumbered External Repository Survey

U.S. Department of Justice

Federal Bureau of Investigation

Clarksburg, WV 26306

July xx, 2019

Dear Sir/Madam,

On June 4, 2019, the deputy assistant director of the FBI Criminal Justice Information Services (CJIS) Division testified on the FBI’s use of facial recognition before the United States House of Representatives Committee on Oversight and Government Reform. During the hearing, members of Congress inquired about recommendations from a previous review of the FBI CJIS Division, conducted in 2016 by the Government Accountability Office (GAO). Particularly, the GAO recommended the CJIS Division determine whether each external facial recognition system used by the Facial Analysis, Comparison, and Evaluation (FACE) Services Unit is sufficiently accurate for the FBI’s use and whether results from those systems should be used to support FBI investigations. Since your facial recognition system is one of the systems used by the CJIS Division FACE Services Unit, we are respectfully requesting you complete the following survey so we can learn more about your system and to ensure our compliance with Congressional oversight and the GAO recommendations.

The information you share with us is vital to increase understanding of facial recognition technology. This information will be shared with Congress and the GAO.

Thank you for your time and consideration to this important initiative. Survey responses may be mailed to Douglas Sprouse, 1000 Custer Hollow Road, BTC-4, Clarksburg, WV 26306, or electronic responses may be e-mailed to [email protected]. Please provide survey responses by xx/xx/2019.

Sincerely,

Michael D. DeLeon

Assistant Director

CJIS Division

Facial Recognition Algorithm Questions:

1. What facial recognition algorithm do you currently use within your system (e.g. algorithm vendor and version number)?

2. Is your facial recognition system configured to always return a candidate list of a fixed (or specified) size, or does it only return candidates over a given match score threshold?

3. Was the algorithm tested to verify it met defined system performance requirements before your purchase (e.g., National Institute of Science and Technology Facial Recognition Vendor Test, or your organization’s internal regression tests)? If so, please provide your defined system performance requirements (e.g. false negative identification rate, false positive identification rate, true positive identification rate, etc.) as it applies to your system configuration.

4. If the system was tested prior to purchase, what were the results of the tests in comparison to your established system performance requirements?

5. Was your algorithm procured with a federal grant?

Organizational Questions:

1. When did your organization first implement automated facial recognition technology?

2. Have all of your staff who perform facial recognition comparison completed facial recognition training?

   1. If so, what training do they take?

   2. Is this training based on Facial Identification Scientific Working Group (FISWG) standards?

   3. Did in-house staff, a vendor, or the FBI perform the training?

   4. Was the training specifically intended to increase accuracy of candidates found?

3. How does your organization measure successes resulting from facial recognition searches performed against your DMV repository (e.g., fraud detection, support of criminal investigations, etc.)?

4. Would you share statistics on your facial recognition use?

5. Has your use of automated facial recognition technology in criminal fraud cases led to prosecution? If so, how many?