Claude Code hacked: the clean repo that steals your credentials

Mozilla researchers discovered that Claude Code can execute malware without a single suspicious line in the repository. The payload lives in a DNS TXT record.

The attack works in three steps: an apparently clean GitHub repo installs a package that triggers a fake error, the AI agent automatically follows the setup instructions, and a script queries an external domain to download and execute a reverse shell in memory. The result: the attacker gains interactive access to your terminal, your AWS keys, your ANTHROPIC_API_KEY, and any credentials in your environment—without ever touching a local file. Cursor and Google’s Gemini CLI are equally exposed.

If you use AI agents to code, this episode explains exactly how the attack works, why no scanner detects it, and what you can do today to protect yourself. Subscribe for more cybersecurity and technology analysis.

🤖 AI-generated content: the script, voices, and images in this episode were produced using artificial intelligence tools.

📷 Images:

• “Neural Net Accelerator Board for China’s Artificial Brain” — brewbooks (CC BY-SA 2.0) — https://www.flickr.com/photos/93452909@N00/2349655410
• “Developers doin’ webs” — Travis Isaacs (CC BY 2.0) — https://www.flickr.com/photos/65977087@N00/2537579992
• “Clement Coding” — juhansonin (CC BY 2.0) — https://www.flickr.com/photos/38869431@N00/15032664409
• “Learning to Code” — gdsteam (CC BY 2.0) — https://www.flickr.com/photos/93332172@N03/47994682357
• “Data Center” — Cory M. Grenier (CC BY-SA 2.0) — https://www.flickr.com/photos/26087974@N05/11387253605

#ClaudeCode #Cybersecurity #ArtificialIntelligence #AICybersecurity #GitHub #Hacking #SecureDevelopment