José Antonio Cordón Muñoz
Online I go by yous. I'm an AI security researcher and engineer at the University of Salamanca, in the BISITE research group — and this is the place where I keep the work, the writing, and the experiments that wander out of the lab.
My work lives where artificial intelligence meets security. I build and study agentic AI systems — autonomous agents, the guardrails that keep their reasoning aligned until a task is actually done, and the new attack surface that large language models open up. Around that sits the rest of it: anomaly detection, threat intelligence, and turning noisy adversarial signal into something a human can act on.
On paper: a computer engineer with a master's in Intelligent Systems — both finished with the University of Salamanca's extraordinary prize — now a PhD candidate in AI at the same university, with 3+ years of research and engineering behind it. My master's thesis, TransformerGuard, built a proactive, self-hosted cybersecurity assistant on open-source LLM agents; it's published, and the fine-tuned model and dataset are open.
But this site isn't the day job. A lot of what I do at the lab stays confidential until it becomes a public paper, so what lands here is deliberately the other side: the things I build in my own time, the ideas I can't stop poking at, a bit of philosophy, the tools and talks, and field notes written as I go.
I don't think an idea is finished until it's running. And in the age of AI-generated code, running is not enough: the real work is in the judgment around it — the framing, the constraints, the prompts, the agents, the evaluations, the automation, and the taste to know when the machine has produced something useful instead of merely plausible. That's the throughline, and the line on the front page: ideas made real, with the judgment the machines still need.
The longer version is scattered on purpose: the systems and talks are in work, the thinking in the blog, the domains I move between in capabilities, and the ways to reach me in contact.
Elsewhere
- GitHub — tools, agent skills and experiments.
- LinkedIn — the professional trail.
- X — occasional notes.
- University of Salamanca — research output.