EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·24d ago·5 min read
Hacking OpenAI
TL;DR
Two critical vulnerabilities were exploited to compromise multiple OpenAI employees' ChatGPT accounts, allowing access to internal repositories. The issue was reported and patched within 72 hours, resulting in a $6,500 bounty.
✦ Why It Matters
Review your own systems for similar vulnerabilities to prevent unauthorized access.
Key Takeaways
- Regularly update software dependencies to mitigate vulnerabilities.
- Implement robust security measures for single sign-on systems.
- Conduct thorough security audits of third-party libraries.
- Stay informed about emerging threats and AI capabilities in security.
How It Works
HacktronAI exploited a heap buffer overflow in the libheif library used by Discourse for image uploads. By manipulating HEIC files, they achieved remote code execution, allowing access to OpenAI's internal systems through compromised accounts.