Indirect Prompt Injection
Everything I do professionally is around helping engineers create amazing applications that are both secure and reliable. That’s why I build engineering tools and educational content that simplify application security.
Throughout my career, I have performed security audits for private and open-source projects, and have found critical vulnerabilities in Google and Mozilla products. I have also taught security to hundreds of engineers and students, while I have also been an external lecturer and Ph.D. candidate in computer science at the Technical University of Denmark.
Here are some of the things I’m working on right now:
- Developing a tool 🛠️ that helps software engineers build applications which comply with privacy requirements
- Creating weekly educational content on application security using comic art 🦇
- Creating a blog 📝 on security at securingbits.com
If you’re interested in learning more about application security, I’d love to hear from you. Feel free to send me a message, and make sure to follow me so I can make security easy for you 🙂
Are you building your next LLM integration? Please consider this:
Integrations for data retrievals can introduce vulnerabilities in your LLM, allowing attackers to inject malicious prompts. This type of vulnerability is known as Indirect Prompt Injection and has been previously illustrated on known models such as GPT4 and Bing.
Learn more about indirect prompt injections from the paper 'Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection' by Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz and Mario Fritz
Link: https://arxiv.org/pdf/2302.12173
