AI use in policing would require disclosure

Summary#

HB 1294, “Use of artificial intelligence-based tools; covered artificial intelligence, disclosure of use,” would amend and reenact an existing section of Virginia law on law-enforcement use of AI. The stated goal is not given in the supplied text; the proposal sets rules for disclosure, records, human decisions, and enforcement. The current version of the law was not supplied, so it is not possible to verify which of these rules would be new.

  • “Covered artificial intelligence” includes AI used to aid a criminal investigation or to write or materially help write police reports and related records. It covers biometric identification, DNA phenotyping, genetic genealogy, cross-camera tracking, vehicle tracking (including automated license plate readers), anomaly detection, gunshot detection, person-based predictive policing, risk scoring, behavioral, sentiment and convoy analysis, fraud and financial-transaction analysis, and social-network or social-media analysis. It excludes administrative tools that do not materially alter an investigation and AI used only for spelling or grammar.
  • Law-enforcement officers would have to disclose covered AI use in the official police report. After an arrest or summons, that report must go to the Commonwealth’s attorney and to the person or their lawyer. Later AI use after arrest must be disclosed to both as soon as practicable and within 30 days.
  • Reports or other investigation records made in whole or part with generative AI would need a disclaimer, identification of AI-generated content where technically feasible, and the author’s certification that they reviewed it for accuracy. The first draft must be kept as long as the final record; the program must keep an audit trail of the user, changes, and any video or audio used.
  • Human decision-makers would remain responsible for decisions on pre-trial release or detention, prosecution, adjudication, sentencing, probation, parole, correctional supervision, and rehabilitation. AI recommendations or predictions remain subject to challenges or objections allowed by existing law.
  • The Attorney General may investigate and sue for equitable or declaratory relief. A resident within an agency’s jurisdiction may also sue for that relief after giving the agency at least 90 days’ written notice to fix the alleged violation. A successful plaintiff is entitled to reasonable attorney fees and costs.

What it means for you#

  • Law-enforcement agencies: They would have to meet the disclosure, recordkeeping, and audit-trail requirements for covered AI use. The bill does not set an application, review, or approval process for using AI.
  • People investigated or charged: After arrest or summons, you or your lawyer would receive the official police report, including the required AI-use details. Later covered AI use after arrest must also be disclosed within 30 days. The bill does not create a new right to challenge AI recommendations; it refers to challenges allowed by existing law.
  • Residents: A person living within an agency’s jurisdiction could bring a civil action to enforce the section, but must first give the agency 90 days’ written notice to cure the alleged violation.
  • People affected by criminal-justice decisions: The listed decisions must be made by a human decision-maker. The bill does not say AI can never inform those decisions.

Money#

No cost information is in the supplied material. No publicly available information.

  • The bill gives no spending estimate, funding source, or staffing number. Agencies would have to meet its recordkeeping and audit-trail rules, but the cost is not stated.
  • An agency could have to pay reasonable attorney fees and costs if a plaintiff prevails. The bill does not set a fine or fee charged to the public.

What is unclear#

  • The current text of the law being amended was not supplied, so what changes from current law cannot be verified.
  • The bill does not set a start date, reporting deadlines for investigations that do not lead to arrest or summons, or a process for giving disclosures in those cases.
  • It does not define how to decide whether AI “materially” aids a report, or what counts as an investigation being materially altered.
  • It does not explain how agencies must ensure an AI program keeps the required audit trail, or what happens if identifying generated content is not technically feasible.
  • It does not set rules for oversight, complaint handling, or remedies beyond the civil actions described.

Case for#

  • The proposal could make AI use in investigations more visible to prosecutors and people facing criminal charges.
  • Keeping drafts and audit trails could help show who used AI, what changed, and what source footage was used.
  • Requiring a human decision-maker preserves human responsibility for major criminal-justice decisions while allowing AI recommendations to be used.
  • The 90-day notice period gives an agency a chance to address an alleged violation before a resident files suit.

Case against#

  • The covered-AI definition reaches many kinds of investigative tools, but the bill does not explain how agencies should classify borderline tools or uses.
  • The disclosure and recordkeeping duties could require agency time and technical changes; the bill gives no cost estimate or funding.
  • Some generated content need only be identified where technically feasible, which could leave gaps in what readers can tell was AI-generated.
  • The bill does not clearly specify when disclosure is due if an investigation does not result in arrest or summons.