Claude Code embeds hidden system prompt markers based on API URL and timezone
By
Thereallo
Summary
A developer inspected the Claude Code binary (version 2.1.196) for privacy and security reasons, discovering that it contains a function which modifies the current date string inserted into the system prompt based on the API base URL and timezone. This suggests the tool is steganographically marking requests with hidden identifiers, raising concerns about transparency in AI coding agents that have extensive system access.
Source
Hacker NewsClaude Code embeds hidden system prompt markers based on API URL and timezonethereallo.devKey quotes
· 3 pulledMost devs give their harnesses ridiculous access. FS, shell, git, browser access, even computer use nowadays.
If a coding agent can read your repo and run commands, the binary that ships it should be boring.
Inside the Claude Code binary, there is a function that changes the current date string inserted into the system prompt.
You might also wanna read
claude-devtools: Open-source tool visualizes hidden Claude Code session data
claude-devtools is an open-source tool that reads raw Claude Code session logs from a user's machine and reconstructs all the information th
Product Hunt·4mo ago
SEO Poisoning Campaign Uses Fake Claude Code Installer to Steal Credentials
Hackers are running an active SEO poisoning campaign targeting users searching for Claude Code installation guides. The attackers create fak
cybersecuritynews.com·1mo ago
Microsoft discovers prompt injection vulnerability in Claude Code GitHub Action exposing CI/CD secrets
Microsoft Threat Intelligence discovered a prompt injection vulnerability in Anthropic's Claude Code GitHub Action that could expose CI/CD w
Claude Code Scheduled Tasks: Automate Recurring Coding Tasks Locally and in the Cloud
Claude Code Scheduled Tasks is a tool that enables developers to run recurring coding tasks across both local desktop and cloud environments
Product Hunt·4mo ago
Mozilla researchers demonstrate indirect prompt injection attack on AI coding agents via GitHub repositories
Mozilla's Zero Day Investigative Network (0DIN) has disclosed a proof-of-concept attack that uses indirect prompt injection to compromise AI
helpnetsecurity.com·5d ago
Sipcode: A context hygiene tool for Claude Code that reduces noise and improves AI response quality
Sipcode is a tool for Claude Code that improves context hygiene by capping verbose tool outputs and deduplicating same-session re-reads, all
Product Hunt·12d ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.