Finally, a Fix for 'Stupid?': Building AI Guardrails for Humans Who Copy-Paste PII

Enterprise AI is having its hair-on-fire moment. Agents are breaking out of their sandboxes. Employees are uploading sensitive data into chatbots at record rates. Public model hubs leak secrets and ship poisoned models. And the regulators all arrived at once — DORA wants your data protected in use, GDPR and HIPAA never left, enforcement for the EU's AI Act is rolling out over the next two years, and the DOJ's new Data Security Program restricts moving Americans' bulk data across borders at all. Everyone's racing to bolt agents onto their stack. Almost no one stops for the first question: is your governance actually alive? Not a policy PDF someone wrote in 2019, not a spreadsheet — a living, breathing control layer that decides, in the moment, who and what is allowed to touch which data. Without that, agentic AI doesn't make you faster. It makes you a breach with a roadmap. This talk is about building that layer — and then doing the fun part safely: letting humans and machines work across your most sensitive data without anyone, or any agent, running amok and spilling it everywhere. I'll show how ontologies, data graphs, and per-request key orchestration decide what each agent may see, catch the queries that would re-identify someone or cross a border, and keep the data encrypted the whole way through. Open-source pieces you can pick up today, and a link to try it yourself. Less data police. More hope: you don't have to choose between using your data and protecting it.