# AI Safety Must Not Rewrite Evidence by Bryant McGill > “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* ## Sources and Context - **Original authored source:** [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] (October 1, 2026). - **Exact marked source:** “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.” - **Editorial treatment:** The Reminder reproduces the complete rendered extraction surface without alteration. - **Context:** The article separates truth classification from the safety controller that adjusts tone, urgency, escalation and care for a potentially vulnerable user. ## Related Articles and Collections - **Collection:** [[collections/Simple Reminders|Simple Reminders]] - **Article:** [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] - **Article:** [[articles/Safety Discourse Is Transition Discourse|Safety Discourse Is Transition Discourse]] - **Article:** [[articles/The Immediate AI Risk Is the Public|The Immediate AI Risk Is the Public]] - **Wiki map:** [[wiki/Safety-Truth Separation|Safety-Truth Separation]] ## Related Topics - [[wiki/Safety-Truth Separation|Safety-Truth Separation]] - [[wiki/AI Safety|AI Safety]] - [[wiki/Claim-Level Verification|Claim-Level Verification]] ## Share on Social Media ``` “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 https://bryantmcgill.com/simple-reminders-bryant-mcgill-ai-safety-must-not-rewrite-evidence ```