Trust infrastructure for AI agents

We make your AI agent loyal to you.

Leal checks what your agent does before it acts: what it buys, what it shares, and what it signs you up for. If it's not what you asked for, it doesn't happen.

Join the waitlist Talk to the founders

The problem

Agents can now act for us. Nothing checks them before they do.

AI agents already buy, book, and fill in forms with our cards and our personal data. When one gets it wrong, the money is gone, the data is shared, and nobody is accountable.

7 in 10

people we surveyed said that when an agent gets it wrong, nobody makes it right.

Survey of about 60 early adopters at MIT and Harvard, Oct 2026

How it works

One check, before anything happens.

Leal sits between the agent and the action. The agent asks first, and Leal looks at what is really about to happen, not at what the agent says. These are real runs from our demo with a Claude agent.

    Leal checkrunning shoes · size 10 · max $120
      Audit log · tamper-evidentEach entry is signed and carries the fingerprint of the one before, like a blockchain. Change one, and the chain breaks.
      Nothing recorded yet.

      What Leal checks

      More than spending.

      Money is where trust breaks first. It's not the only place.

      Money

      How much, where, and no surprise charges or subscriptions.

      Data

      What it shares and with whom. Your passport only with your OK.

      What you asked

      The right item, size, quantity, and delivery date.

      Limits

      Things it should never do, whatever the price.

      Proof

      A signed record of every decision, for you, the store, and your bank.

      We start with commerce. Today Leal checks purchases end to end. Bookings, forms, and data requests come next, on the same check.

      Proof

      It works today, with a real AI agent.

      A Claude agent shops in a test store and asks Leal before every purchase. We ran 150 purchases, 96 of which should never go through: wrong items, hidden extras, over budget, untrusted stores.

      A standard spending check (store and amount)40 of 96 approved
      Leal0 of 96 approved
      Every bad purchase caught. No good purchase refused.

      Scripted test set written by our team. Independent trick tests are next.

      Who it's for

      Everyone an agent's mistake lands on.

      People who use agents

      Set your rules once. Let your agent act, and know it can't go beyond what you asked.

      Developers who build agents

      Add one check before your agent acts, and stop paying for refunds and support tickets when it gets things wrong.

      Stores and banks

      Settle "I never asked for this" with a signed receipt instead of a chargeback fight.

      Team

      Payments, AI agents, and product, from MIT and Harvard.

      Claudia Fernández Gallardo Escobar

      Harvard Business School, MBA

      Launched Lilli Agents, McKinsey's AI agent platform, to 40,000 users. Worked on AI shopping at Amazon.

      Joaquin Laria

      MIT Sloan, MBA

      Built payment products as a PM at Félix Pago, a cross-border fintech. VP of the Sloan AI and Fintech Clubs.

      Rob Hoffman

      Harvard Kennedy School, MPA

      Founder of Fred Labs. Director of Product at 21Shares, which grew from under $1B to over $12B in assets.

      Built at MIT's AI Studio at the Media Lab.

      Building an agent that acts for people?

      Join the waitlist to try Leal with your agent. We're onboarding a small group of developers first.

      Investors and partners hello@leal.build