Most outlets assert they're neutral. Apprised has a rival AI model (Kimi) independently grade every desk brief for bias risk — political lean, source-diversity skew, loaded framing — before it publishes, and we publish the running record, including the briefs that got flagged. Last 30 days, updated 2026-06-12.
Share of each desk's audited briefs rated LOW. Desks with too few audited briefs yet are shown as building.
When the audit doesn't clear a brief as LOW, it records why. The recurring flags in this window:
Before each desk brief publishes, an independent model that did not write it (Kimi) reads a condensed version and rates its bias risk LOW / MODERATE / HIGH with specific flags. This ledger is the deterministic tally of those verdicts — no new judgment is added here. We surface the MODERATE and HIGH counts and the flags prominently on purpose: a 100%-clean record would mean the audit isn't doing its job. Real MODERATE readings (with flags) are in the record above. The published editorial voice is still Claude; this is a second model checking it. See the full methodology and our forecasting track record.
https://apprised.news/index/bias-ledger
Live JSON: /api/bias-ledger. Each brief carries its own "Bias-reviewed" pill that aggregates here.