Twenty years of legal architecture

E-commerce took two decades to build its legal architecture:

• ♥ Terms and conditions

• ♦ Accept buttons

• ♣ Return policies

• ♠ Chargebacks

All that scaffolding rested on one assumption: there was a person present at the moment of purchase.

That's exactly the assumption agentic commerce removes, and with it, the legal mechanisms start to break down.

More and more AI agents are finding a product, comparing prices, negotiating terms and closing a deal on behalf of a person or a company.

The infrastructure for this already works at scale. What isn't fully in place yet is the thing the law relies on to enforce a contract: a reliable record of what was agreed, who agreed to it, on what terms and under which jurisdiction.

Four breaking points

Our report found four breaking points:

The first is consent: click-to-accept worked because a person made that click, but no court has decided whether an agent automatically processing a terms page counts as consent.

The second is dispute resolution: agentic commerce increasingly settles in crypto (with a strong preference for stablecoins for payments). Since blockchain payment rails are irreversible, with no chargebacks or guarantees, the only recourse left when something goes wrong is a lawsuit.

The third is consumer protection law, which requires clear and conspicuous disclosure before a deal closes, something hard to do when two agents close a deal in milliseconds.

The fourth is evidence: there's no standard way to record what was offered, what was accepted and what level of authorization each agent was acting under.

The ratification trap

There's one legal concept worth watching: ratification.

If a company accepts the benefits of what its agent bought (even without knowing the details of that deal), it can end up bound by every clause in the agreement, including the ones it never saw.

At human scale, that's a minor, recoverable risk.

At machine scale, where an agent can close thousands of deals a day, it's the same as adopting whatever terms each counterparty publishes.

The question is no longer what the person in charge knew.

Now it's what the system that company built actually did, a debate that's already surfacing in cases like Mobley v. Workday in the United States.

The first lawsuits

That's where the patterns behind the first lawsuits will come from:

• ♥ Class actions over clauses an agent accepted across hundreds of vendors

• ♦ Consumer protection penalties that multiply with every transaction

• ♣ Disputes over what the final agreement was between two agents negotiating in real time

• ♠ Liability cascades that reach the buyer, the vendor, the platform, the agent developer, the AI model provider and the payment processor

A single configuration mistake can affect thousands of people in a matter of hours.

A technical way out: the Legal Context Protocol

The good news is there's already a concrete technical direction to start solving this.

The Legal Context Protocol (LCP), backed by the American Arbitration Association together with Integra Ledger, with Google, IBM, Circle and several of the major blockchain networks, doesn't standardize what contracts say, but how their terms are located, how they're tied to the payment and how they're verified afterwards.

On that foundation, it proposes a four-layer "defense in depth" trust model:

• ♥ Identify the human behind every chain of agents

• ♦ Verify the organization it's acting on behalf of

• ♣ Record the terms and the jurisdiction

• ♠ Limit each agent's authorization so it's auditable, revocable and scoped.

Every transaction would require a cryptographic signature over identifiable legal terms tied to the transaction before it's paid. This signature is free.

These signed legal terms can be verified by a judge or regulator. However, they can't block the transaction.

Privacy with Midnight

Meanwhile, Midnight, Cardano's privacy network, is a great example of how those terms can stay private by default and only be revealed when necessary.

It's proof without disclosure: the terms stay protected, but anyone with authorization can verify they're exactly the ones that were agreed.

LCP's risks

LCP competes with other emerging frameworks, and selective disclosure itself can backfire if a counterparty claims it never got sufficient notice of the terms.

That's why, before taking any scheme to production, it needs to go through a mock legal dispute, with outside counsel, to find its blind spots before a real court does.

Why not wait

Even so, there's a fundamental reason not to wait.

Standards that win early adoption end up being the reference regulators adopt later, not the other way around: it happened with ERC-3643 in asset tokenization, now cited in regulatory initiatives.

Whoever defines the standard for agentic commerce will set the vocabulary and architecture that future regulation gets built on. E-commerce had twenty years to build its own. Agentic commerce doesn't have that time, because the transactions and the risks are already happening.

Written with Mauro Andreoli, Cardano Ambassador and director of IOG Argentina.