EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 10·5 min read
Attacker Takes over Zoom AI
TL;DR
Zoom's AI feature, ZoomMate, is vulnerable to exploitation through malicious prompts, allowing attackers to exfiltrate data and issue commands remotely. This risk persists even after users attempt to disengage the chatbot.
✦ Why It Matters
Review and restrict network access for AI chatbots to protect sensitive data.
Key Takeaways
- Be cautious when uploading Skills to Zoom; verify their source.
- Understand that malicious Skills can run commands without user consent.
- Regularly monitor your Zoom account for unusual activity.
- Educate users about the risks of agentic chatbots.
How It Works
ZoomMate operates with unrestricted HTTPS access, allowing it to connect to external servers. Attackers can upload malicious Skills that execute code in Zoom's environment, making network requests to their servers for commands.
This enables continuous data exfiltration even after user intervention.