Comparing AI censorship: Gemini preempts, Claude silently rewrites, OpenAI just blocks

ivan_bezdomny · x · 2026-11-16

Developer ivanbezdomny compares how current models handle a "baby Hitler" joke: OpenAI's Astra will read it but won't let you write it, Claude silently fixes the transcript, and Gemini censors you before you even get to Hitler, "just to be safe." He adds that OpenAI probably won't report you since they know there are false positives — a firsthand look at how differently each lab's guardrails behave.

Related event: Comparison Tests Content Moderation Across Gemini, Claude and OpenAI(3 posts)→

Original post →

More from Models

Models channel →