Recent AI Security Insights

Hey friends! 😊 

I hope you’re doing well, especially all the new faces!

This email list grew from 200 to OVER 1200 in the last two weeks!! It had me looking at the blog, and realizing that I haven’t posted there since the end of November, so I’ll make time to write on there soon.

In the mean time, I’ve been actively sharing and posting great stuff on Twitter/X, so I thought I’d outline some of that here. This is effectively a roundup from the end of 2023 and the first few weeks of 2024.

What You (May Have) Missed The Last Two Months

🙈 INVISIBLE PROMPT INJECTION was found by Riley Goodside. I did a breakdown on Twitter and a video on youtube about it. I’m way too calm in the video for how exciting the finding is. I’ll bring the energy in future videos.

📔 NIST released an amazing publication on AI Security. I covered some of the highlights in this Twitter thread.

🛡️ META’s LLM GUARD: In the middle of December, META released an amazing paper on their AI Safety protection mechanism. The “model incorporates a safety risk taxonomy.” It’s well worth a read for those interested in the safety and security of AI-Human interactions. Link: HERE.

❓ WHAT AI FRAMEWORK TO USE: I have been asked “what AI framework should I use to build X?” so many times recently that I wrote a reference tweet I could copy and paste. I think you’ll love it.

👨🏽‍💻 WEB LLM ATTACK LABS by PORTSWIGGER: Portswigger has some of the best labs for learning and practicing application security. They released some LLM-based ones with real-world practical appsec vulns. They’re amazing. Go check them out: https://portswigger.net/web-security/llm-attacks 

🎥 FREE AI SECURITY CONFERENCE TALK: This was in my last email, but as there are 1,000 new subscribers, I am going to mention it again. I did a talk on AI Application Security—mostly prompt injection—for an online conference in December. I put the recording on YouTube for all of you here.

👩🏼‍💻 AI FILESYSTEM is a really cool minimal way to search for data on your file system with embeddings. I love the simplicity, implementation, and the way it could be leveraged by so many projects/ideas. Check it out here.

Thanks for being on the email list! 😊 If you like this content, I’d love if you invited someone to join it or to follow me.