Adrian Wedd

Systems builder, AI safety researcher, and adversarial thinker — writing code since age six, with fifteen years of applied infrastructure, security, and delivery across government, university, and campaign organisations. The failure-first methodology didn't come from papers — it came from coordinating communications and logistics for Greenpeace's Actions unit against well-resourced opponents where getting the risk assessment wrong had real consequences. That operational security instinct runs through everything: leading the cybersecurity uplift for Tasmania's public housing sector — translating the ~900-control ACSC ISM into a tracked delivery program alongside Essential Eight and IDAM work; an adversarial evaluation practice spanning 257 models and 142,068 adversarial prompts across 346 attack techniques; and production AI systems with hallucination detection and content safety gates built in from the start. Currently freelancing across client delivery (healthcare ecosystems, multilingual sites, ops tooling) and independent AI safety research. AuDHD — the hyperfocus and pattern recognition are features, not bugs. Most interested in where complex systems break and what that reveals about how they were built.