H.R. 9477: AI Incident Reporting Thresholds

Overview

H.R. 9477, an introduced AI Incident Reporting Act, would delegate coverage thresholds and important implementation choices to the Secretary of Commerce. This series follows the proposed path from covered models and developers through reportability, timing, escalation, under-detection, and system design, then offers a threshold-first test for future rules without treating the bill or the research framework as settled law.

Learning outcomes

H.R. 9477 is an introduced proposal, not enacted law; its operational coverage thresholds remain for future rulemaking.
A reporting form cannot repair missing data, undefined harm categories, or triggers that ignore cumulative and ongoing conditions.
Pre-harm, near-miss, aggregate, and tolerance-based paths fit different risks and require different evidence.
A credible regime would need protected disclosure, cross-source visibility, validation, and feedback—not simply a longer incident list.

Keywords

H.R. 9477reportingthresholdscoveragereportabilityescalationcausationaggregationbaselinesoversight

Chapters

H.R. 9477: Coverage Before Reporting

Why the boundary comes before the form

12 min · Why the boundary comes before the form

12 min

H.R. 9477: Covered Models and Developers

How the Secretary’s thresholds define the perimeter

9 min · How the Secretary’s thresholds define the perimeter

9 min

H.R. 9477: Reportability and Timing

Why early signals need context and supplements

11 min · Why early signals need context and supplements

11 min

H.R. 9477: Escalation Beyond Severity

How data, definitions, and triggers shape action

14 min · How data, definitions, and triggers shape action

14 min

H.R. 9477: Threshold Blind Spots

How timing, aggregation, and vulnerability create blind spots

11 min · How timing, aggregation, and vulnerability create blind spots

11 min

H.R. 9477: A Reporting System That Learns

How definitions, incentives, and validation close gaps

10 min · How definitions, incentives, and validation close gaps

10 min

H.R. 9477: The Threshold-First Test

How to evaluate coverage, evidence, triggers, and blind spots

9 min · How to evaluate coverage, evidence, triggers, and blind spots

9 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-7 Premium

FAQ

What does H.R. 9477 propose AI developers report?

H.R. 9477 would require a covered model developer to report activity it knows or reasonably believes is reportable, including control-undermining behavior, model-weight compromise, specified dangerous capabilities, and some fortuity-prevented serious risks.

Why do thresholds matter more than an AI incident-reporting form?

Thresholds decide which models and developers are covered, what signals count, how much evidence is enough, and whether the response is ordinary or urgent; a form can only record those choices.

How could an AI incident system detect cumulative or ongoing harm?

It would need a defensible unit of analysis, shared harm categories, continuous data, and aggregation or tolerance-based monitoring; the research paper presents these as design proposals, not requirements already established by H.R. 9477.

Does current evidence show that H.R. 9477 would reliably detect AI incidents?

No. H.R. 9477 is introduced and leaves key thresholds unresolved, while the research paper’s ten-incident walkthrough identifies under-detection mechanisms but does not establish universal reliability.