Detect and Deny (D2) - Simple Deterministic Guardrails for LLM/Agent

Hey Product Hunt - Cybersecurity founder here. I have spent the past couple of years breaking AI agents through prompt injection. Once you control the agent, the game is over. I've exfiltrated data, chained together function calls the agent should never make, and bypassed every nondeterministic safety check thrown at me. The problem is that most teams are trying to secure AI agents with more AI. They use LLMs to judge whether actions are safe. But those judgments are probabilistic, same input, different output. You can't build real security on top of that. My cofounder and I have built deterministic guardrails because authorization decisions need to be predictable and verifiable. Not "the model thinks this is probably okay" but "this user with this role calling this function with these parameters - allow or deny, every single time." What makes this different is that it works at the function level where agents actually operate. Role-based access control on individual functions. Input validation that catches malicious parameters before execution. Sequence enforcement that prevents multi-step attacks like blocking an agent from reading your database and then calling an external API in the same session. The biggest attack pattern I've seen is agents being tricked into multi-step exfiltration. Deterministic sequence enforcement stops that cold. This is what I wish existed when I was doing agent security research. Happy to talk through specific attack scenarios if anyone's interested https://calendly.com/artoodavid/...