Tyche Institute · LIMEN edge-case atlas

Forum-reach league

Which kinds of AI harm actually reach a regulator or a court, and which stay stuck in media-only reporting? This table ranks the 16 harm categories in the ObscureAI/LIMEN dataset by forum reach — the share of documented cases in each category that carry a regulator or court record, rather than only a news report.

What "forum reach" means

Loading…

Forum reach by AI-harm type (gradient chart)

Bar chart: forum reach percentage by AI-harm category, solid bars for categories with a pre-existing mandated forum, outline bars for novel machine-actor harms with no dedicated forum.

Harm category	Cases (n)	Forum reach	Media-only	Distinct sources	Reading

Inferential rows (multi-forum covered) are categories where cases are documented across several independent source lanes (regulators, courts, press, security advisories), so the reach percentage is reasonably comparable across categories. Descriptive-only rows come from a narrower source mix (often dominated by a single lane) — read the reach percentage as a description of what that lane happened to catch, not a comparable estimate of true forum reach. Categories with fewer than 10 cases are additionally flagged small n because a handful of cases can swing the percentage sharply.

Source: papers/attributability-debt-atlas/analysis-2026-07-04/results.json (RQ1_by_category, RQ2_forum_fit), N=1,044 — the same case pool behind the public ObscureAI dataset.