Expired50% confidenceFactTime unknown
The 'undercover mode' instructions in the leaked code require Claude to write commit messages like a human developer, not mention that it is an AI, not reference Claude Code, not include internal model names, and not leave any co-attribution that reveals AI involvement.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Anthropic Source Code Leak: Unreleased Model Codenames and Hidden Features Deep Dive
bilibili量子菠萝_
Related Claims
UnverifiedThe GetUndercoverInstruction function instructs Claude to write commit messages like a human developer, without mentioning an AI identity, without mentioning Claude Code, and without including internal model names.83% similarUnverifiedClaude Code is classified as an AI Agent, not a traditional Q&A-style large language model.72% similarUnverifiedClaude Code的设计定位是可编程AI代理平台，而非通用对话助手71% similarUnverifiedClaude Code's source code contains a "Stealth Mode" that intentionally makes generated code variable names rough and removes comments, in order to prevent AI-generated code from contaminating training data.71% similarUnverifiedClaude Code SDK provides programmatic access to Claude Code Agent capabilities in headless mode71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/60131API
curl https://kongchang.com/api/v1/knowledge/claims/60131MCP
get_claim(id=60131)