WHAT THIS BILL REGULATES · 2 REQUIREMENT TYPES
How Is This Bill Enforced
Verbatim statutory text on the left; plain-language analysis and a per-section checklist on the right. Numbered markers cross-link to the matching checklist row.
This Act may be cited as the ''Justice in Forensic Algorithms Act of 2021''.
This section establishes the short title of the Act as the "Justice in Forensic Algorithms Act of 2021." It creates no compliance obligations.
(a) 1 Not later than 1 year after the date of enactment of this Act, the Director of the National Institute of Standards and Technology shall establish a program to provide for creation and maintenance of standards for testing computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1), to be known as the Computational Forensic Algorithm Testing Standards, consistent with the following:
(a)(1) 1 Testing standards shall include an assessment for the potential for disparate impact, on the basis of race, ethnicity, socioeconomic status, gender, and other demographic features.
(a)(2) 1 Testing standards shall address— (A) the underlying scientific principles and methods implemented in computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1); and (B) requirements for testing the software including the conditions under which it needs to be tested, types of testing data to be used, testing environments, testing methodologies, and system performance statistics required to be reported including— (i) accuracy, including false positive and false negative error rates; (ii) precision; (iii) reproducibility; (iv) robustness; (v) sensitivity; and (vi) system failure rates; (C) requirements for publicly available documentation by developers of computational forensic software of the purpose and function of the software, the development process, including source and description of data used to develop the tool, and internal testing methodology and results, including source and description of testing data; (D) requirements for laboratories and any other entities using computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) to validate it for use, including to specify the conditions under which the lab has validated it for their use, requirements for what information needs to be included in a public report on the lab or other entity's validation, and requirements for internal validation updates when there are material changesMaterial changeThe term "material change" means an update to computational forensic software that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.Section 2(h)(2) to the software; and (E) requirements for reports provided to defendants by prosecution produced documentation by the use and results of computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) used in individual cases.
(a)(3) 1 Testing standards shall be issued as a rulemaking under section 553 of title 5, United States Code.
(a)(4) 1 The Director shall consult with outside experts in forensic science, bioethics, algorithmic discrimination, data privacy, racial justice, criminal justice reform, exonerations, and other relevant areas of expertise identified through public input.
This subsection directs the Director of NIST to establish, within one year of enactment, the Computational Forensic Algorithm Testing Standards — a comprehensive set of testing requirements for computational forensic software. The standards must address disparate impact across race, ethnicity, socioeconomic status, gender, and other demographic features, the underlying scientific principles of the software, detailed testing methodology requirements (covering accuracy, precision, reproducibility, robustness, sensitivity, and failure rates), developer documentation requirements, laboratory validation requirements, and requirements for prosecution reports provided to defendants.
The standards must be issued through notice-and-comment rulemaking under 5 U.S.C. § 553, and the Director must consult with outside experts in forensic science, bioethics, algorithmic discrimination, data privacy, racial justice, criminal justice reform, and exonerations.
(b)(1) 2 There shall be no trade secret evidentiary privilege to withhold relevant evidence in criminal proceedings in the United States courts.
(b)(2) Nothing in this section may be construed to alter the standard operation of the Federal Rules of Criminal Procedure, or the Federal Rules of Evidence, as such rules would function in the absence of an evidentiary privilege.
This subsection eliminates the trade secret evidentiary privilege in federal criminal proceedings. It prohibits withholding relevant evidence in criminal cases on the basis of trade secrets, while preserving the standard operation of the Federal Rules of Criminal Procedure and the Federal Rules of Evidence in the absence of such a privilege. This is a significant due-process protection for criminal defendants facing evidence produced by proprietary forensic algorithms.
(c) 3 Any Federal law enforcement agency or crime laboratory providing services to a Federal law enforcement agency using computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) may use only software that has been tested under the National Institute of Standards and Technology's Computational Forensic Algorithm Testing Program and shall conduct an internal validation according to the requirements outlined in the Computational Forensic Algorithm Testing Standards and make the results publicly available. The internal validation shall be updated when there is a material changeMaterial changeThe term "material change" means an update to computational forensic software that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.Section 2(h)(2) in the software that triggers a retesting by the Computational Forensic Algorithm Testing Program.
This subsection restricts federal law enforcement agencies and their affiliated crime laboratories to using only computational forensic software that has been tested under NIST's Computational Forensic Algorithm Testing Program. Agencies must additionally conduct an internal validation under the Computational Forensic Algorithm Testing Standards and make the results publicly available. Internal validations must be updated whenever a material software change triggers retesting by the NIST program.
(d) 4 The Director of the National Institute of Standards and Technology shall establish a Computational Forensic Algorithm Testing Program, whose activities include the following: (1) Testing individual software programs using the testing requirements established in the Computational Forensic Algorithm Testing Standards. (2) Using realistic sample testing data similar to what would be used by law enforcement in criminal investigations in performing such testing, including incomplete and contaminated samples. (3) Using testing data that represents diversity of racial, ethnic, and gender identities and intersections of these identities in performing such testing. (4) Using testing data that tests the limits of the software and demonstrates the boundaries of reliability described in the performance measures defined in the Computational Forensic Algorithm Testing Standards in performing such testing. (5) Publishing the results of testing the software online including results under conditions specified in the testing standards and across diversity of racial, ethnic, and gender identities and intersections of these identities in a publicly available format.
This subsection directs the NIST Director to establish the Computational Forensic Algorithm Testing Program, specifying its required activities: testing individual software programs against the standards, using realistic and diverse testing data (including incomplete and contaminated samples and data representing racial, ethnic, and gender diversity and intersections), stress-testing software reliability boundaries, and publicly publishing all test results disaggregated by demographic identity categories.
(e) 5 Retesting shall be conducted when a material changeMaterial changeThe term "material change" means an update to computational forensic software that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.Section 2(h)(2) is made to the software that impacts its performance and may affect its outputs. The Director shall establish requirements for determining whether changes are material or nonmaterial.
This subsection establishes the trigger for retesting: computational forensic software must be retested whenever a material change is made that impacts performance and may affect outputs. The NIST Director must establish requirements for distinguishing material from nonmaterial changes. This creates an ongoing monitoring obligation tied to the definitions in Section 2(h).
(f) 6 Any results or reports resulting from analysis by computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) shall be provided to the defendant, and the defendant shall be accorded access to both an executable copy of and the source code for the version of the computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1)—as well as earlier versions of the software, necessary instructions for use and interpretation of the results, and relevant files and data—used for analysis in the case and suitable for testing purposes. Such a report on the results shall include— (1) the name of the company that developed the software; (2) the name of the lab where test was run; (3) the version of the software that was used; (4) the dates of the most recent changes to the software and record of changes made, including any bugs found in the software and what was done to address those bugs; (5) documentation of procedures followed based on procedures outlined in internal validation; (6) documentation of conditions under which software was used relative to the conditions under which software was tested; and (7) any other information specified by the Director of the National Institute of Standards and Technology in the Computational Forensic Algorithm Testing Standards.
This subsection imposes disclosure obligations on the prosecution's use of computational forensic software in individual cases. All results and reports from forensic software analysis must be provided to the defendant, along with both an executable copy and the source code for the software version used (including earlier versions), instructions for use and interpretation, and all relevant files and data. The report itself must include specified elements: the developer's name, the lab where the test was run, the software version, change records including bugs, documentation of procedures followed, documentation of testing-vs.-use conditions, and any additional information NIST specifies.
(g) 7 In any criminal case, evidence that is the result of analysis by computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) is admissible only if— (1) the computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) used has been submitted to the Computational Forensic Algorithm Testing Program of the Director of the National Institute of Standards and Technology and there have been no material changesMaterial changeThe term "material change" means an update to computational forensic software that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.Section 2(h)(2) to that software since it was last tested; and (2) the developers and users of the computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) agree to waive any and all legal claims against the defense or any member of its team for the purposes of the defense analyzing or testing the computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1).
This subsection creates an evidentiary admissibility gate for computational forensic software results in federal criminal cases. Evidence produced by such software is admissible only if two conditions are met: (1) the software has been submitted to and tested under NIST's Computational Forensic Algorithm Testing Program with no material changes since its last testing, and (2) the developers and users of the software agree to waive all legal claims against the defense team arising from the defense's analysis or testing of the software. This provision provides the enforcement mechanism for the bill's testing and transparency requirements by making non-compliant forensic evidence excludable.
(h)(1) The term ''computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1)'' means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.
(h)(2) The term ''material changeMaterial changeThe term "material change" means an update to computational forensic software that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.Section 2(h)(2)'' means an update to computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) that may affect the performance measures defined in the Computational Forensic Algorithm Testing Standards or the use or output of the software.
(h)(3) The term ''nonmaterial changeNonmaterial changeThe term "nonmaterial change" means an update to computational forensic software that does not affect the performance measures, use, or output of the software.Section 2(h)(3)'' means an update to computational forensic softwareComputational forensic softwareThe term "computational forensic software" means software that relies on an automated or semiautomated computational process, including one derived from machine learning, statistics, or other data processing or artificial intelligence techniques, to process, analyze, or interpret evidence.Section 2(h)(1) that does not affect the performance measures, use, or output of the software.
This subsection provides the bill's three defined terms: computational forensic software (the covered platform), material change (the retesting trigger), and nonmaterial change (updates that do not trigger retesting). The computational forensic software definition is notably broad, covering any software relying on automated or semiautomated computational processes — including machine learning, statistics, or other AI techniques — to process, analyze, or interpret evidence.