# Claim-Level Verification **Entity class:** AI verification method **Domain:** Artificial intelligence / calibration / evidence **Maturity:** Emerging research architecture ## Definition **Claim-level verification** decomposes a generated response into atomic propositions and evaluates the support, contradiction, uncertainty and confidence of each proposition instead of assigning one undifferentiated confidence score to the entire answer. ## Mechanism The method separates claim decomposition, evidence retrieval, evidence evaluation, risk estimation and hallucination localization. It is especially important when one fluent paragraph combines verified statements with unsupported causal claims. ## Article Context [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] argues that claim-level verification should occur before accommodation, theology, mental-health framing, political sensitivity and stylistic generation. ## Evidence Boundary Claim-level confidence does not guarantee source quality or causal correctness. The result still depends on retrieval coverage, evidence adjudication, contradiction handling and calibrated abstention. ## Relationships - [[wiki/Universal Epistemic Engine|Universal Epistemic Engine]] - [[wiki/Evidence Sufficiency|Evidence Sufficiency]] - [[wiki/Model Calibration|Model Calibration]] - [[wiki/Epistemic Uncertainty|Epistemic Uncertainty]] - [[wiki/Ground Truth|Ground Truth]] ## Simple Reminders, Quotations, and Thoughts > “Safety systems in conversational AI may govern the urgency, tone, and care of a response, but they must not govern whether evidence is classified as support, contradiction, or uncertainty when the system evaluates descriptions of reality against what is known to exist.” > **— Bryant McGill**, *Epistemic Invariance in Conversational AI, 2026* [[reminders/AI Control/AI Safety Must Not Rewrite Evidence by Bryant McGill|AI Safety Must Not Rewrite Evidence by Bryant McGill]] ## Sources / Provenance - [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] — architectural placement and research synthesis. - Abbasli et al., “Claim-Level Confidence Calibration for Reliable Decision Making with Large Language Models” (2026), as cited in the article.