Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Algorithmic Tools topic
No spam. Unsubscribe anytime.
ACUS adopts recommendation for agency disclosure and oversight of algorithmic enforcement tools
Summary
The Administrative Conference adopted a recommendation urging agencies to assess, disclose, and mitigate risks when using algorithmic tools for regulatory enforcement, narrowing the scope to exclude basic scientific tools and widely used software and adding data-category disclosure requirements.
The Administrative Conference of the United States on Thursday adopted a recommendation urging federal agencies to adopt stronger transparency, risk assessment, and public‑engagement practices when deploying algorithmic tools in enforcement work.
The recommendation, from ACUS’s Committee on Regulation, asks agencies to evaluate harms and benefits early in a tool’s lifecycle, plan exit ramps if a system performs poorly, and give the public a means to review categories of data used to train or drive the tool. "Risk assessments typically don't deal well with individualized harms," consultant Michael Karanaklis told the assembly as he summarized the report. He called for safeguards that include "appropriate exit ramps, and planning for a system to be retired if it fails to perform." (Michael Karanaklis, consultant.)
During debate, members pushed on the recommendation’s definition of "algorithmic tools," arguing a broad definition could sweep in routine agency software. Committee chair Helen Seracio (spelling in transcript: Seracios/Seracio) said the recommendation was "really about making sure if agencies are going to be using algorithms in enforcement, they should understand what the pros and cons are" and emphasized transparency and public comment. (Helen Seracio, Committee on Regulation chair.)
The assembly adopted a narrowed formulation after several council and member amendments. The final text keeps a working definition — "a computer‑based process that uses rules or inferences drawn from data to transform specified inputs into outputs to make decisions or support decision‑making" — and places clarifying language in a footnote to exclude "basic scientific tools" and "widely used computer software." Members also added an explicit expectation that agencies disclose the categories and sources of data used by tools, so potential issues of bias can be identified and assessed.
The assembly voted to adopt the recommendation by voice and unanimous consent following the amendment process. The chair said the language aims to balance agencies’ need to use data‑driven tools with protections for due process and public confidence in enforcement.
What happens next: ACUS intends the recommendation to guide agencies and provide model practices; the assembly’s staff asked members to send any drafting comments on related statements or drafts by Jan. 5 so the office can finalize texts for publication.
