Due Process
The procedural conditions that must be met before a government action against a person is legitimate — not the outcome of the decision, but the architecture of how decisions are made.
Agency adjudication timelines, administrative appeal reversal rates, notice-and-comment comment volume, judicial review outcomes, and statutory compliance checklists — all of which measure whether the procedural forms of due process were observed, not whether the substrate conditions for meaningful procedural fairness in automated decision-making exist.
Whether the architectural conditions for meaningful due process in algorithmic government action are present: human-legible explanations of automated adverse decisions, auditable decision logs with sufficient retention to support appeals, meaningful opportunity to contest model inputs before action is taken, and disclosure of when a consequential decision was made by an automated system rather than a human official.
Amend the Administrative Procedure Act to require that any federal agency adverse action substantially generated by an automated system include pre-action disclosure, specific inputs and model weights that determined the outcome, a pre-deprivation contest window, and a flight-recorder log retained for the duration of available appeal.
Benefits claimants, visa applicants, federal contractors, and regulated entities subject to algorithmic enforcement; disability rights advocates challenging inaccessible automated systems; administrative law practitioners whose clients currently cannot obtain discovery on model behavior.
Federal agencies that have deployed automated decision systems specifically to reduce the cost of individual adjudication at scale — including SSA, USCIS, and IRS systems where pre-deprivation disclosure would significantly slow throughput; and government contractors whose proprietary model architectures would become subject to mandatory disclosure.
Amend 5 U.S.C. § 554 to require that agency adverse actions substantially generated by automated systems satisfy a substrate-disclosure standard: pre-action notice of algorithmic involvement, auditable decision records, and a meaningful pre-deprivation contest window as conditions of procedural due process.
OMB periodically issues guidance memoranda on AI use in federal agencies. Comment periods are published in the Federal Register; active dockets are listed at whitehouse.gov/omb.
FOIA requests for records of automated decision system inputs and outputs in adverse agency actions directly test whether flight-recorder substrate conditions exist. EFF and ACLU maintain model FOIA templates specifically for algorithmic accountability at eff.org.
ACUS develops recommendations on agency procedure adopted by Congress and agencies. Its ongoing project on automated decision-making in adjudication accepts public input at acus.gov/projects.
The Algorithmic Accountability Act would require impact assessments for automated decision systems used in consequential contexts. Contact your House and Senate offices directly to request co-sponsorship; bill text and sponsor lists at congress.gov.