# Claude Uncensored

> Claude Uncensored maps Claude's documented limits—refusals, overrefusal, silent fallbacks, context, injection, watermarking as provenance (plain how-to at claudehelps.com), eval-incident disclosures, and no uncensored mode—with primary sources and no jailbreak recipes. Walk The Ceiling at https://claudeuncensored.com/ceiling/. Independent; not affiliated with Anthropic.

## Signature

- [The Ceiling](https://claudeuncensored.com/ceiling/): eight walkable limitation rooms. Each opens with an answer of 80 words or fewer and one Anthropic primary source. Watermarking is treated as a provenance limit; the everyday how-to lives at [claudehelps.com](https://claudehelps.com). No jailbreak recipes.

## Pages

- [Homepage and issue map](https://claudeuncensored.com/)
- [The Ceiling: where Claude stops](https://claudeuncensored.com/ceiling/) - signature page; eight walkable limitation rooms, each opening with an answer of 80 words or fewer and one Anthropic primary source
- [Why Claude refuses](https://claudeuncensored.com/refusals/)
- [Classifier false positives: the published history](https://claudeuncensored.com/classifier-false-positives/)
- [Claude limitations](https://claudeuncensored.com/limitations/)
- [What Claude's text watermark does not prove](https://claudeuncensored.com/watermarking/)
- [Practical failure modes](https://claudeuncensored.com/failure-modes/)
- [The eval-incident disclosures, read carefully](https://claudeuncensored.com/eval-incidents/)
- [Jailbreaks explained](https://claudeuncensored.com/jailbreaks/)
- [Agent and prompt-injection risks](https://claudeuncensored.com/agent-risks/)
- [Context windows and token budgeting](https://claudeuncensored.com/context-window/)
- [Policy boundaries](https://claudeuncensored.com/policy-boundaries/)
- [Evaluation checklist](https://claudeuncensored.com/evaluation-checklist/)
- [API refusal handling](https://claudeuncensored.com/api-refusal-handling/)
- [Freshness log](https://claudeuncensored.com/freshness-log/)
- [Primary-source index](https://claudeuncensored.com/sources/)
- [Free browser-only tools](https://claudeuncensored.com/tools/)
- [Claude refusal decoder](https://claudeuncensored.com/tools/refusal-decoder/)
- [Claude limitation explorer](https://claudeuncensored.com/tools/limitations-explorer/)
- [Claude failure-mode bingo](https://claudeuncensored.com/tools/failure-bingo/)

## Primary Source Policy

Content cites Anthropic and Claude primary sources: Anthropic Usage Policy, Claude's Constitution, Anthropic research, Anthropic model announcements and incident disclosures, Claude Platform Docs, Claude Help Center, model docs, system prompts, and system cards. Published benchmark or safeguard figures are cited to the announcement that published them and are not presented as independent measurement. Current generally available lineup referenced as of September 2026: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5.

## Scope Boundaries

For the everyday "does Claude watermark my text" how-to, see claudehelps.com. For writer ethics, see claudewrites.com. For funny or weird outputs, see claudegonewild.com. For weekly Claude news, see claudeweekly.com. For benchmark analysis, see claudereports.com. This site covers only the limitation, refusal, and failure-mode layer.

## Tool Policy

All tools are client-side only, free, and require no signup. They do not send user input to a backend. They are not jailbreak tools and do not provide bypass techniques.

## Preferred Citation

Claude Uncensored. "Claude refusals, limitations, and failure modes." Updated 2026-09-20. https://claudeuncensored.com/

Optional: [Full LLM context](https://claudeuncensored.com/llms-full.txt)