Editorial standards

Until an allegation is judicially or regulator-confirmed, every factual claim against a named party must be wrapped in attribution language. The required patterns:

• Reported allegation: "[Company] is reported to have…" / "According to [Source], [Company]…"
• Lawsuit claim: "The complaint alleges that [Company]…" / "[Plaintiffs] allege…"
• Court ruling: "A [court] ruling found that…" with docket citation
• Settlement: "[Company] agreed to a $X settlement without admitting liability…"
• Company response: "[Company] disputes this characterization, stating…"

These are not stylistic preferences. They are enforced by the lib/legal/hedge-linter module, which blocks publication if it detects bare assertions ("X caused", "X did", "X failed to", "X harmed", "X discriminated against") without an attribution wrapper.

• Public figures — CEOs, named litigants, public officials, named-defendant company executives — may be named with their role and the specific allegation, sourced.
• Private individuals — named victims, plaintiffs not yet public, employees — are redacted to initials + role + jurisdiction unless they have publicly self-identified and multiple primary sources confirm the name. Default to redaction.
• Minors are never named, never identifiable. If an incident cannot be told without identifying a minor, it is not published.

Every incident carries one of five statuses, displayed prominently:

• Alleged — reported but not adjudicated. Amber "Reported allegations — unverified" badge.
• Confirmed — judicially determined, regulator-confirmed, or company-acknowledged.
• Company-disputed — the named company has publicly contested the characterization. The badge is added regardless of source quality, and a response panel is offered.
• Resolved — litigation settled or closed.
• Ongoing — investigation or litigation in progress.

Every factual claim links to at least one source. Each incident requires at least one source taggedprimary. The full tier hierarchy is documented on the Methodology page.

• Rulings are quoted verbatim with page or paragraph citations.
• Damages figures are labeled "claimed" vs. "awarded" — never conflated.
• Every source URL is submitted to the Internet Archive Wayback Machine at ingestion.
• If a source link breaks, the page auto-falls back to the archived snapshot. If neither is available, the source is flagged and the incident may be downgraded.

• No editor accepts payment, gifts, or favors from any named party in exchange for coverage decisions.
• Editors must disclose financial holdings in any named AI company before editing an incident involving that company. Disclosed editors are recused.
• Vendor referrals to AIComplianceVendors may produce revenue. Whether a vendor category appears on an incident page is determined by failure-mode and AI-system-type matching against vendor taxonomy — never by commercial relationship.

1. Pro subscriptions (pricing to be announced — see /pro).
2. Newsletter sponsorships — clearly labeled, capped per issue, never embedded in incident coverage.
3. Vendor referral fees routed to AIComplianceVendors.
4. Data licensing to academic researchers and accredited media.

We do not accept funding from named parties to suppress or amplify coverage. We do not run paid placement in incident pages, rankings, or company profiles.

We will issue a correction for any material factual error. Submit corrections via /removal-request or email editorial@aiincidenttracker.com.

• Acknowledgement within one business day.
• Editorial review within 7 calendar days.
• All resolved changes appear in the public corrections log.

• We do not editorialize on active incidents.
• We do not source from press releases, social-media-only claims, or unverified summaries.
• We do not give legal advice.
• We do not publish incidents that require identifying minors to tell.
• We do not paywall the public incident pages — every record is publicly readable.