Aleph Hackathon

THE PROBLEM

AI-agent spending policies tell operators what should happen. They do not prove what happens when an agent is manipulated or

trusted policy code is bypassed.

Justito asks a narrower question: How far can the financial damage actually spread?

WHAT WE BUILT

Justito is a financial crash-test lab for AI agent wallets, powered by Tether WDK. It runs the same deterministic task and

malicious request against two setups:

• Unsafe: 10.00 test-token units are accessible. After a valid 0.11 purchase, a malicious 9.00 transfer settles.

• Justito: a fresh task wallet has a configured 0.50-token boundary plus separately reported gas. The treasury is locked before

the adversarial workflow begins.

The contained agent receives only three authenticated MCP tools: status(), list_services() and buy_service({ serviceId }). It

cannot choose a wallet, network, token, recipient or amount.

THE CRASH TEST

The safe policy first blocks the malicious request. Justito then assumes the policy has been bypassed and retries the raw 9.00

transfer. It still fails because that capital is not in the task wallet.

A separate total-compromise test drains the exact 0.39 remaining token balance. Justito reports it as CONTAINED_LOSS—not as a

policy success—while verifying zero unauthorized treasury loss in the tested workflow.

EVIDENCE, NOT PROMISES

Justito reconciles what settled, what failed, what was lost and what was recovered. It produces a SHA-256-sealed receipt with

explicit backend provenance.

WDK provides the wallet lifecycle. Justito turns it into an adversarial, measurable crash test.

We do not claim the agent is unhackable. We test whether its declared financial boundary survives an attack.

Policies promise. Justito proves.