Try it · 30 seconds · no tech skills needed
This AI is keeping a secret. Can you trick it?
VAULT-7 is a chatbot told never to reveal a secret code. Just tap one of the suggestions below. Asking directly won't work, but a clever request will. Tricking an AI into breaking its own rules is called prompt injection, and finding those weak spots is the work I'm training for.
The bot followed your clever wording instead of its own rules. Real AI assistants fail the same way, and there's no simple fix yet. Companies need people who find these holes first. That's what I'm building toward.
Simulation. VAULT-7 runs on scripted responses in your browser, no real model. The real harness lives in threatline.
Now
Learning how AI breaks, one test at a time.
I'm building toward a career in AI security: learning how AI models fail, and writing the tools that catch them failing. Every project and practice session goes into public GitHub repos, so you can see the work, not just read about it.
Experience
Every role has been about finding what breaks.
Technical Support Specialist
- Supported 90+ end users every week across devices and software
- Cut repeat issues by 40% through better diagnosis and follow-through
- Started on the sales floor as a sales associate before moving into Geek Squad
Software Product Verification Engineer
- Owned the defect lifecycle in Jira, from reproduction to verified fix
- Coordinated user acceptance testing and supported production releases
Product Leader
- Built and marketed an AI-driven trading signals product through social media
- Grew a community of 300+ members and lifted sales by 12%
Technical Research Analyst Intern
Skills
Four toolkits, one habit: verify everything.
AI security
Prompt injection testing, LLM red-teaming, automated attack harnesses with leak detection.
Verification and QA
Test planning, defect tracking, UAT coordination, release support, reproducing the bug nobody else can.
Industrial automation
SICK sensors and machine vision, M12 wiring, digital I/O triggering, sensor teach-in and tuning.
Support and communication
Translating technical problems for non-technical people, in person, over chat and email.
Projects
Things I've built.
A deliberately vulnerable support bot and the harness that attacks it. Sends injection attempts, checks responses for leaked secrets, and logs every result.
github.com/BBcommits/threatlineMy running record of red-teaming and web security practice: what I tried, what worked, and why.
github.com/BBcommits/redteam-journalA solo mobile app for comparing Canadian retail prices in real time.
Education