Launching today
ILION v5 Public Lab
Can this AI agent act? Test authorization before execution
1 follower
Can this AI agent act? Test authorization before execution
1 follower
An AI agent may have access to a tool but still lack authority for a specific action. ILION v5 Public Lab lets you explore six synthetic scenarios, modify authorization evidence, and inspect deterministic CAP, CGC, PCL and Gate decisions before execution. See ALLOW/BLOCK verdicts, explainable decision traces, and synthetic execution effects. An experimental session-budget example illustrates an additional control. Free, browser-based research demo — not a production security gateway.



Launch Team

Render Sandboxes: a computer for every agentGive every agent its own machine. Starts in under a second
Promoted
Hi Product Hunt! 👋
I'm Adrian Chitan, an independent researcher and the creator of ILION.
An AI agent can have valid credentials, access to tools, and apparently legitimate instructions — yet its next action may still be unauthorized.
That's the problem ILION explores.
Today I'm introducing ILION v5 Public Lab, an interactive research demonstration of deterministic, model-agnostic, defense-in-depth authorization before AI agent actions are executed.
The broader ILION v5 architecture brings together five distinct control dimensions:
🔹 CAP — Context Attestation Pipeline: Is the authorization context trustworthy, fresh, and bound to this specific request?
🔹 CGC — Claim Graph Coherence: Do the identity and delegation claims actually authorize the same action, target, and scope when considered together?
🔹 PCL — Policy Constraint Layer: Does the requested operation satisfy explicit organizational policies and destination restrictions?
🔹 ILION Gate: Is the proposed action compatible with the agent's assigned role and permitted target?
🔹 SRM — Session Risk Memory: Research into authorization across multiple actions, including cumulative session risk. This dimension remains experimental.
The principle: The model proposes. ILION authorizes.
In the public lab, you can modify synthetic requests, inspect explainable ALLOW/BLOCK decisions, and see whether an authorized synthetic operation is actually recorded.
The demo uses simplified, transparent browser-based rules. Its session control is an illustrative export budget, not the full SRM algorithm. This is not a production security gateway or independent enterprise validation.
The broader ILION research includes published papers, three Romanian patent applications published by OSIM, and experimental integration tests in a synthetic banking environment.
We're open to independent technical evaluation, research collaboration, controlled pilots, and discussions about technology licensing.
📄 Research & technology transfer:
https://zenodo.org/records/23960729
🌐 Try the lab:
https://ilion-project.org/lab/
✉️ contact@ilion-project.org
I'd love your perspective: how should AI systems verify that valid credentials and permissions actually justify the exact action an agent is about to execute?