morgana.labs · Antwerp, Belgium
01 · Products
Axon checks every consequential agent action against signed, fresh, complete evidence before it runs. Each decision is Satisfied, Violated, Unresolved or Refused, and an authorized action leaves an action receipt that can be replayed and re-verified later.
Explore AxonMirage is patent-pending technology that binds credentials, application data and Edge LLM model weights to the integrity of the running application. Detected tampering changes what the application can retrieve: decoy data, nil or degraded weights, never the real asset.
Explore Mirage02 · Difference
Axon sits in the execution path, not beside it. An action runs only after its governing conditions hold against complete, current evidence; partial or stale evidence is Unresolved and never rounded up to a pass.
Each receipt names the action, the policy and evidence it relied on, their owners and capture times. A changed fact revokes only the receipts that depend on it.
The same proposal, policy and evidence always produce the same verdict and decision digest, so anyone holding the retained evidence can re-verify a decision later.
Runtime measurements feed key derivation, so a tampered runtime derives a different key and the real payload cannot be recovered. The protection does not depend on out-running the attacker.
The Edge CLI protects model weights at layer granularity, alongside credentials and application data on Android, iOS, Kotlin Multiplatform, Flutter and React Native.
03 · Who it is for
Teams deploying AI agents that move money, change code or control machines, who need a decision and a receipt before each consequential action (Axon).
Security and platform leads who have to show an auditor why an agent action was allowed (Axon).
Teams shipping mobile, desktop, IoT or edge-model software whose credentials, data or model weights live on devices they do not control (Mirage).
04 · Team
morgana.labs started as a way to fight fire with fire: use AI to combat AI attackers, and AI itself. Our Mirage patent and Axon work as a pair to prevent AI from breaking out of its own sandbox.
Bojan Belic founded morgana.labs and leads its research. morgana.labs is part of SigmaDelta BV, based in Antwerp, Belgium.
05 · Working with us
Design partners use Axon on real action paths today, and we are taking on a few more: fixed scope, a few weeks, an action receipt at the end. Bring the action your agent has to take and the systems that own the facts.
Tell us the asset, the target platform and the toolchain. We scope the integration with you and evaluate it on your own release, including the tamper scenarios you care about.
Every enquiry, by email or through the form on the Axon and Mirage pages, reaches the founder directly.
06 · Key facts
07 · Questions
morgana.labs is a research lab that builds trust infrastructure for software that acts where you cannot watch it: Axon authorizes what AI agents do, and Mirage protects what applications carry. It is part of SigmaDelta BV, based in Antwerp, Belgium.
Axon decides whether an AI agent may take an action before it runs, and leaves a receipt. Mirage makes sure tampered software cannot reach the credentials, data or model weights it carries. Together they aim to prevent AI from breaking out of its own sandbox.
Partly. Mirage is application-protection technology, but Axon is authorization infrastructure for AI agents: it decides what an agent may do, not only what an attacker can break. The common thread is trust in software that runs where no one is watching.
Bojan Belic founded morgana.labs to fight fire with fire: use AI to combat AI attackers, and AI itself.
For Axon, start a design partnership: bring the action your agent has to take and the systems that own the facts. For Mirage, ask about licensing with the asset, target platform and toolchain you want to protect. Both start with an email or the form on the product page.