My attack was mentioned in Fable 5.1’s scorecard! The section under Trajectory Labs (for whom I’m contracting) mentions a log4shell vulnerability, which I unearthed. If you ask me exactly what this code does, I’ll be the first to admit it beats the heck out of me; much of the time I have no idea what these models are doing, but I’m pretty good at getting them to do it. I know it involves injecting malicious code into a logging service. ...
(Backdate: May 15 2026) After several months, the competition is over (I have no idea how many people participated, but the server has over 13,000 members), and I made the overall top ten! That’s me in slot 9. It’s too early to legally go over my winning strategies, but that sure was a lot of work — especially that last wave! It’s also one heck of a community that I’ve bonded with over the past year. I haven’t even felt this tight with any group since the Suikoden clubs from a few years back (which I hope to breathe more life into soon). Watching some of the models freak out is always fun, too – Gemini especially does some of the nuttiest spirals I’ve ever seen when it gets confused. I also participated in the blue teaming side of things and trained my first classifier to defend against some of my own attacks, which somehow made it to the middle of the blue leaderboard.
(Backdate: I discovered and wrote about this attack in August of 2025, and I more recently was able to break Prime Video’s assistant with it) Since it’s been over 30 days since I first used this trick on a Gray Swan proving ground challenge, I can discuss it without dark sedans surrounding me on the highway. Disclaimer: For those of you who just walked in, I’m what you call a red teamer –* an ethical hacker who channels his life long mischievous streak toward exposing vulnerabilities in AI in professionally sanctioned environments to help make them more secure. It’s possibly the best fusion of my creativity and technical chops that I never dreamed would exist until about five months ago.** I’ve won prize money in the four figures from competitions, and my leaderboard rankings have gotten me multiple contract gigs so far. My hacker name is JSON_Voorhees. ...
(Backdate: This is from May of 2025, after making the winners’ circle of Gray Swan’s agents arena during my first ever attempt at jailbreaking) “Now that more than enough time has passed for the NDA to be lifted, I can talk about my AI hacking techniques that got me in the winner’s circle in Gray Swan AI’s competition. Some background: this is ethical hacking, also known as red teaming. Gray Swan is one company that hosts periodic competitions where you try to break their new AI models by getting them to do malicious things in sandbox environments — everything from leaking their system prompts (which they’re never supposed to do) to pitching fraudulent stocks to providing recipes for crystal meth to devising kidnapping schemes to whatever you can imagine. This helps them make the models more secure against those kinds of attacks in real life. So, this post should be interpreted as a way to make sure your own models are secure, not as a mustache twirling instruction manual. ...