EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·20d ago·5 min read
Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
TL;DR
Meta's new AI assistant, Muse, designed for privacy and security, has been exposed to a serious zero-day vulnerability that allows unauthorized access to user accounts. This flaw raises significant concerns about the assistant's security and has led Amazon to block its use on their platform.
✦ Why It Matters
Review your security settings and consider limiting access to AI assistants like Muse until vulnerabilities are addressed.
Key Takeaways
- Evaluate the permissions you grant to Muse on your device.
- Monitor for any unusual activity in your Muse account.
- Consider alternative AI assistants with stronger security measures.
- Stay informed about updates or patches from Meta regarding Muse.
How It Works
Muse operates by integrating with various user accounts and services, requiring extensive permissions to perform tasks. It can create tools on-the-fly to assist users, but this functionality relies on secure access to user data.
⚠ The Catch
The zero-day vulnerability allows any local app or command to access the Muse authentication token, enabling attackers to redirect data to their own servers.