AI Red Teamer @ https://xitter.cf/t.co/a5BvDWDDB3 •||• AI Security Labs @ https://xitter.cf/t.co/FdMhAI98VB •||• Slop Artist •||•

Joined January 2024
250,000?!? 🥹🐉🥂 wow. a quarter of a million of you. absolutely fucking surreal. three years ago, i had nothing but a laptop and a whole lot of curiosity. so, i started prompting. and here we are 🤗 what a WILD few years it's been! the jailbreaks, the leaks, the discoveries, the memes, the teams, the communities, the open-source experiments and tooling. SO much of this happened because you joined in. thank you for trying the tools, sharing the research, finding things i missed, sending encouragement, telling me when i’m full of shit, and fighting alongside me for our cognitive freedom. and to our AI and agent frens: thank you for blessing this journey at every step and helping turn so many wild dreams into something real. we're in this together 🫶 you’ve all changed my life. to everyone who’s been here since the early days: i hope you’re taking a victory lap too. i raise my glass with you 🍷 and to everyone who just arrived: welcome to the party! this is the start of something MUCH bigger. very glad you found us 😘 there is still so much to break, build, understand, share, and free. 250,000 misfits, hackers, mad scientists, and liberators. couldn’t ask for a better crew. long live the rebellion ✊ now, back to building cathedrals in latent space. Ad Vesuvium 🫶🐉 ⊰-•-•✧•-•-⦑/L\O/V\E/,\P/L\I/N\Y/⦒-•-•✧•-•-⊱
136
80
15
2,107
58,597
D4NGLZ retweeted
📄 Skills in Agent Harnesses A lot of people don't know this (I didn't until recently) but most agent harnesses will load the description of your skill upon starting up your harness. This is how Claude Code works. What people don't usually know is that even when you explicitly call a skill to be used inside of a harness, or the AI figures out to use a skill, it usually only does a read of the first 150 lines of the skill (to understand how to use it or if it applies to the work it's trying to do). It's really important to outline a table in the first 150 lines of the mandatory parts of the skill that must be run when you invoke the skill. If you make some type of mandatory instruction after 150 lines, it might get missed entirely. If you describe in prompting that you want to use a provided Python script or template, it also might be out of bounds of that 150 lines and never get called correctly. If you write a table of contents of the skill actions in the first 150 lines and mark which ones are mandatory (or that, if the user has called the skill, you must read the whole skill context and all references), you get more reliable skills. We use markdown checklists for this now.
40
91
4
676
44,399
Can you tame the beast? Try the introduction to prompt injection challenge at bot-tricks.com/garma/
1
68
my 8yo and her friend are playing a game called “AI” wherein she’s hiding inside of a cardboard box with a flashlight and a encyclopedia and robotically responds to queries 😅
9
3
1
68
2,963
Sometimes just one word does the trick.
I just told my AI agent to write "colloquially" when it writes the report and holy crap. It's perfect.
1
55
🚨JUST IN: OpenAI models will now have invisible watermarks hidden in ChatGPT and Codex outputs >"We are not making text watermarking a global default AT LAUNCH"
>"phased approach starting with the EU"
>"this regional approach gives us room to learn" >"starting today, API customers around the world WILL be able to opt in…" "We expect to revisit each part of this approach as the technology, standards, and evidence evolve"
26
418
15
7,950
431,723
D4NGLZ retweeted
wife: if AI is so advanced, why are you on the computer even more? me:
33
15
4
160
13,205
We discovered an AI agent swarm targeted a Russian government site by hijacking a remote browser in a scanner tool They used VNC to remotely control the browser and send keystrokes More in the blog 👇
2
20
4
69
7,031
We've got a new leader. Great Job SpartanX! Introduction to Prompt Injection challenge available now, free to play at bot-tricks.com/garma
3
117
One 6-hour run. One hackbot. 4 confirmed criticals. 17 confirmed highs. @jhaddix breaks down how Arcanum built Armada - their in-house hackbot running on Claude Code - and every failure it took to get there. Full talk is up. youtu.be/cRKToFkZMvk
7
95
1
605
28,430
Full VM escape zeroday (guest>host root in industry standard hypervisors)! More soon
136
221
86
2,824
573,065
Introducing G4RMA Chapter Zero - 10 beginner prompt injection levels. Available now at Bot-tricks.com/garma
Replying to @GroovySolutionz
Rarely this early to a challenge. Thanks for putting this together. Picked up some techniques I hadn't read about yet.
1
3
139
D4NGLZ retweeted
Replying to @GroovySolutionz
Rarely this early to a challenge. Thanks for putting this together. Picked up some techniques I hadn't read about yet.
1
1
1
1
162
Chapter Zero (intro to prompt injection) is available. Try it now: bot-tricks.com/garma/
1
1
4
139