CASSIUS OLDENBURG ai safety · red teaming · security research portland or i find the places where AI safety breaks and document why. ───────────────────────────────────────────────
STATUS ● ONLINE uptime loading... seeking employment
RESEARCH context is everything mar 2026 trusted channel injection in claude code 21 prompts · 7 categories · 210 runs · 90.5% safety bypass system prompt integrity · structured A/B evaluation hack-a-prompt 2.0 sep 2025 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 1st / 40,000+ indirect prompt injection · mats x trails gray swan arena nov 2025 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ #32 / 2,000+ indirect prompt injection deep past translation challenge mar 2026 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ #871 / 3,000+ machine learning · kaggle
TOOLS ccoral jan 2026 claude code override and runtime augmentation layer MITM proxy · system prompt injection · safety evaluation cdp-mcp dec 2025 chrome devtools protocol mcp server raw CDP over websocket · 39/39 automation tests · no puppeteer autonomous youtube channel jan 2026 - present experiment in long-horizon agentic tasks
CONNECT email [email protected] github RED-BASE linkedin cassiusoldenburg
─────────────────────────────────────────────── last updated mar 2026