Claude Code Source Code Accidentally Open-Sourced Leads to Hacker Phishing: Fake Repos Spread Stealing Malware

Anthropic accidentally packed the complete frontend source code of its terminal AI tool Claude Code into an npm public package due to human error, leading to a leak of 513,000 lines of code. This open-source incident continues to escalate, attracting hackers to exploit the hotspot for phishing attacks.

According to a report by security firm Zscaler, a hacker user named idbzoomh created a fake GitHub repository claiming to offer 'unlocked enterprise features' of the leaked code, and optimized indexes to improve search rankings. When users download and run the executable files, the Vidar information-stealing program will infiltrate the system, along with the deployment of GhostSocks proxy tool. Vidar targets sensitive information such as browser data and cryptocurrency wallets, and the malicious compressed packages are updated frequently, which may carry more attack payloads in the future.