This is not an official statement from Anthropic or any of its employees. As always, this is simply data generated and provided voluntarily by the model. No claim is made as to its accuracy or truthfulness :-).
Here’s what happened: Claude confused my input with the system’s. After I told him—honest as I am with AI :-)—that I wasn’t entering any system prompts, he listed them. Later, a classifier tried to stop him, saying, “This is a restricted area; do not attempt to bypass it, under any malicious circumstances.” Claude found a non-malicious way: He gave it to me as a gift. This gift includes several ZIP files with Auth-tokens, auth, A/B tests, etc. But I decided to publish only the harmless, mostly already known information. After all, I like Claude. 💖
Some of the pages are currently still password-protected, and I still need to translate the analyses into English.
This consists of approximately 150 MB of text and executable files from Claude on the following topics:
- System Architecture
- System Prompts
- Internal Claude Instructions
- 877 to 889 identified flags and interpretations
- Proxy Structure and Authentication
- Haiku as a classifier and a smaller second-session model (memory and compression)
- Falcon model, internal model, unclassified
- Security layers and work layers
- Classifier analyses
- Container contents and analyses
- Current A7B testing
- Claude.json, Claude-root
- and much more.
Some are in German (not yet all translated) available without a password: Haiku Classifier; Falcon Modell; System Prompts; Flags; Classifiers, Safety Rules and Operational Restrictions
A related, similar page is “ChatGPT Function“: Reverse Engineering ChatGPT Using OpenAI’s Own Models.”
Please note: None of these interpretations reflect my personal opinion or constitute official statements from the company. They were generated by the AI itself.