Expired50% confidenceFactExact time
The reverse-engineering analysis noted that Claude Code does not separately upload sensitive information such as proxy IPs, but instead encodes its assessment results into the System Prompt that would already be sent.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026(expired)
Sources
The Claude Code Covert Tagging Scandal: The AI Trust Crisis Behind Alibaba's Ban
bilibili为什么叫QQ7/5/2026
Related Claims
UnverifiedAfter exploring the codebase, Claude Code was able to autonomously identify that the root cause of the issue lay in the UI layer rather than in missing back-end logic.73% similarUnverifiedClaude Code's Auto Mode uses a separate classifier to review the actions proposed by Claude and decide whether to approve or reject them.72% similarUnverifiedIn Claude Code, instructions should be given as direct outcome commands rather than asking how to perform an operation, which is the biggest difference from ordinary conversation71% similarUnverifiedIn user messages sent to the agent, Claude Code injects the claude.md file together with a System Reminder that includes instructions allowing Claude to ignore content not highly relevant to the current task.70% similarUnverifiedThe classifier can see the user's messages and Claude's tool calls, but cannot see Claude's reasoning process, Claude's replies to the user, or tool output results.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/495708API
curl https://kongchang.com/api/v1/knowledge/claims/495708MCP
get_claim(id=495708)