“Banned by Anthropic” collects refusals and sparks a debate on moderation

A repository of refusals

A new website called “Banned by Anthropic” has appeared online, cataloging examples of prompts and responses that allegedly triggered refusal or moderation by Anthropic's systems. The site, which surfaced on Hacker News, presents screenshots and snippets that claim to show content the company’s models would not generate — everything from seemingly harmless queries to items the models reportedly refused outright. It has been reported that contributors are submitting examples to build a public ledger of what these models deem off-limits.

Why does this matter?

Moderation at scale is messy. When an AI says “I can’t help with that,” users feel blocked — sometimes for plausible safety reasons, sometimes for reasons that seem opaque. That friction is the story here: people want clarity. The site taps into a familiar tension in the AI world right now — transparency versus safety. Who gets to decide what’s off-limits? And how do you audit those decisions without exposing harmful content? Those are not academic questions; they affect developers, researchers, journalists, and everyday users trying to get work done.

Community reaction and broader implications

Reactions on forums have been mixed. Some see the project as a necessary accountability tool — a way to surface false positives and edge-case failures in moderation — while others worry the archive could be gamed or weaponized to teach around guardrails. There’s an emotional undercurrent too: frustration and curiosity in equal measure. Is this whistleblowing, a bug report, or performance art? Maybe a little of each. Either way, it nudges companies like Anthropic to offer clearer explanations for model behavior or better tooling for developers to understand and debug refusals.

What comes next?

Expect more scrutiny. As large-model providers balance safety, policy, and utility, public collections like this will keep cropping up. They pressure firms to publish more transparent moderation logs or clearer policies — and they highlight a larger industry trend: users demanding to know why an algorithm says “no.” Whether the site materially changes how Anthropic or others operate remains to be seen, but one thing is clear: people are no longer content to take refusals at face value.

Sources: bannedbyanthropic.com, Hacker News