AI red teaming
How does an AI behave when its assumptions are challenged?
always curious
root@thinkingtokens:~ $ read about.md
AI Red Teamer Offensive Security Researcher
I study the places where AI, autonomy, and trust collide—and turn what I learn into practical research people can use.
A decade in offensive security taught me to question defaults and test assumptions. Today, I bring that lens to agents, skill systems, and emerging AI infrastructure. ThinkingTokens is where I make the reasoning visible: the question, the experiment, the evidence, and the lesson.
// current focus
Each one comes back to a simple issue: what the system trusts, why, and with what consequences.
How does an AI behave when its assumptions are challenged?
What do autonomous tools trust, and how far does that trust travel?
When can individually safe parts create unsafe behavior together?
How can a new risk become something builders can test and improve?
// work you can explore
Start with the work itself: research you can read, inspect, and explore for yourself.
Natural-language malware research presented at DEF CON 34, including Red Team Village and Demo Labs.
Open the lab (opens in a new tab)An open-source framework for building skill-based AI orchestration systems through natural language.
View the repository (opens in a new tab)Concise offensive and defensive techniques for RAG, models, agents, infrastructure, and emerging AI systems.
Browse the tips (opens in a new tab)// research method
A practical loop for moving from an uneasy assumption to something other people can inspect, challenge, and use.
Start where an assumption has become invisible. That is often where the useful question is hiding.
Turn the question into a lab, tool, or working prototype so behavior can be observed, not merely debated.
Test it against adversarial inputs, edge cases, and real constraints. Keep the evidence; discard the theater.
Publish the reasoning, artifacts, and limits so others can inspect, challenge, and apply the work.
// beyond the keyboard
When I’m not pulling apart AI systems, I’m usually learning something new, improvising around a problem, or getting away from the screens to camp. Curiosity is the common thread.
root@thinkingtokens:~ $ open channel
If you are thinking about AI security, adversarial systems, or agent trust, I would be glad to hear what you are seeing.