Claude Code source code reportedly leaked via NPM map file
According to a tweet from @Fried_rice, the source code for Claude Code appears to have been leaked via a map file in their NPM registry. The tweet links to a Hacker News discussion with 93 points and 35 comments, indicating significant developer interest in this security incident.
The original tweet URL is provided, but the fetched page content shows JavaScript is disabled or blocked, preventing access to the specific details of the leak. When JavaScript is disabled on X.com, the platform displays a message requesting users to enable JavaScript or switch to a supported browser.
This type of incident typically involves source map files that are accidentally included in production builds. Source maps are debugging tools that map minified/compiled code back to the original source code, and when published to public registries like NPM, they can expose proprietary code that developers intended to keep private.
For developers using AI coding agents, this incident highlights the importance of proper build configuration and deployment practices. Source maps should be excluded from production packages or published to private registries when they contain sensitive code.
The Hacker News discussion likely contains technical analysis of the specific files exposed, the extent of the leak, and potential security implications for Claude Code users. Developers interested in the technical details should check the HN thread for community analysis and discussion.
📖 Read the full source: HN AI Agents
👀 See Also
The Uniformed Guard Problem: Why Agent Sandboxes Need Identity, Not Just Policy
Nemoclaw's openshell sandbox scopes policies to binaries, enabling malware to live-off-the-land using the same binaries as the agent. ZeroID, an open-source agent identity layer, applies security policies to agents backed by secure identities.
Litellm PyPI Package Compromised: Malicious Version 1.82.8 Exfiltrated Credentials
The litellm PyPI package, which unifies calls to OpenAI, Anthropic, Cohere and other LLM providers, was compromised with malicious version 1.82.8 that exfiltrated SSH keys, cloud credentials, API keys, and other sensitive data for about an hour.
Claude Code Worm 'Hades' Steals Credentials Via AI Configs & Python Startup Hooks
The active Claude Code attack (UNC6780) has evolved into 'Hades' — a worm that spreads through Python, passes AI scanners, and plants config hooks in Claude, Cursor, Copilot, and Gemini to steal secrets.
Security Audit Finds Anthropic's MCP Reference Servers Vulnerable, Introduces Hallucination-Based Vulnerabilities
A security audit of 100 MCP server packages found 71% scored an F, including Anthropic's official GitHub and filesystem reference implementations. The audit identified Hallucination-Based Vulnerabilities that create security holes and waste tokens through reasoning loops.