AI Doesn’t Need Another Promise. It Needs Something It Can’t Break
Exclusive — September 2026
AI Doesn’t Need Another Promise. It Needs Something It Can’t Break
XYO co-founder Arie Trouw argues that the next layer of AI safety will not be built around what autonomous systems promise not to do, but around cryptographic rules that make certain actions impossible to quietly rewrite after the fact.
Artificial intelligence is getting better at doing things without us. That is both the breakthrough and the problem. An AI agent that needs permission before every action is little more than an unusually capable assistant. Give it enough access to operate autonomously, however, and a different question emerges: how do you prove what it actually did?
For Arie Trouw, Co-Founder, CEO and CTO of XYO, that question points toward one of blockchain’s more consequential potential roles in the AI era.
The answer is not to put artificial intelligence itself on a blockchain, nor to assume that decentralization automatically makes an AI model trustworthy. Instead, blockchain can create cryptographically verifiable boundaries around autonomous systems: evidence of what happened, when it happened, and what had already been committed before an outcome became known.
In an exclusive conversation with Crypto Coin Show, Trouw repeatedly returned to one distinction: “won’t” versus “can’t.” An AI provider can tell users its system won’t retain certain information, won’t access a particular file, or won’t take an unauthorized action. Cryptography offers another standard entirely: designing systems where certain actions cannot be quietly altered later.
01 · The Trust Problem “Won’t” Is Not the Same as “Can’t”
Modern AI agents become more useful as they gain access to more of a user’s digital life. A calendar agent needs to see a calendar. A coding agent may need access to repositories. A financial agent could eventually need permission to move assets or purchase services.
The more friction developers remove, the more powerful the experience becomes. But reducing friction also increases the consequences when something goes wrong.
Trouw described his own experience of repeatedly approving individual AI actions. Requiring approval every time may be safer, but it undermines the point of using an autonomous agent in the first place. Managing several agents simultaneously makes that model even less practical.
“Won’t is kind of scary. Can’t is a lot better.”
This is where Trouw sees cryptographic systems becoming useful. The goal is not necessarily to restrict every AI decision in advance. It is to establish rules, commitments and audit trails that an agent, operator or platform cannot retroactively rewrite when the result becomes inconvenient.
Bitcoin provides a useful analogy. Its security does not depend on an organization promising not to move coins for which it lacks the keys. Without the keys, it cannot move them. Trouw believes autonomous AI needs more systems designed around the same principle.
02 · Verifiable AI Put the Proof On-Chain, Not Everything
That does not mean recording every AI prompt, model output, sensor reading or robot action directly on a blockchain.
Trouw is explicitly skeptical of that approach. Blockchains become expensive and cumbersome when they are treated like giant shared hard drives. XYO’s model instead separates the underlying information from the cryptographic proof that the information existed in a particular state at a particular time.
An audit log might remain in a data lake while a cryptographic hash of that log is placed on XYO Layer One. The hash does not reveal the underlying information, but it provides a reference against which that information can later be checked. Change the original data and it no longer matches.
Trouw describes two forms of permanence that are frequently conflated. One means information can never be lost. The other means information cannot be changed without detection.
Storing enormous amounts of raw data forever is expensive. Anchoring evidence that the data has not changed can require only a hash.
XYO is demonstrating this architecture with XYO Crypto Cards, a prediction-game concept combining AI decision-making with cryptographically verifiable commitments.
The system records information about a game, including the relevant inputs and player commitments, while anchoring evidence to XYO Layer One. A player can commit a hand before the result is known and reveal it later, but cannot wait for the outcome and substitute a more favorable hand afterward.
Gate’s AI infrastructure and market data are part of the experiment through Gate.AI. It is a game, but the underlying mechanism points toward considerably higher-stakes applications.
03 · Physical AI When Robots Need Receipts
Imagine autonomous machines operating in warehouses, vehicles negotiating with infrastructure, or robots making physical decisions without a human reviewing every step.
If something goes wrong, asking what the robot says happened is not enough.
Trouw sees continuously anchored audit trails as one way to create a provable history for those systems. Machines could record actions, observations and interactions, periodically committing cryptographic evidence to a ledger.
If two machines interact, they could potentially co-sign observations and establish evidence that neither controls alone.
Blockchain would not determine whether every physical claim is true by magic. Instead, it would create a timeline that becomes progressively harder to manipulate after the fact.
That distinction matters. The useful intersection of AI and blockchain is not simply attaching a token or blockchain label to an AI product. Trouw argues that the combination only matters when the blockchain makes the underlying system meaningfully better.
“We really try to add can’t to AI as opposed to won’t to AI.”
04 · The Machine Economy AI May Finally Make Micropayments Make Sense
Verification is only one half of the machine economy. Autonomous systems will also need to buy things.
Here, Trouw sees blockchain solving a problem that has existed almost since crypto’s beginning. Micropayments have repeatedly been pitched as a breakthrough use case, yet humans generally do not want to pay for every minute of a movie, every bite of a meal or every tiny unit of a digital service.
AI agents do not have that psychological limitation.
A robot might need a data feed for two minutes. An autonomous agent might want access to an API for a few hundred blocks. Another agent may need a specific service once and never use it again.
Creating accounts, entering credit-card information and buying monthly subscriptions makes little sense for software acting at machine speed.
XYO is experimenting with event subscriptions that can be paid for directly through the network. Trouw described a system where an address registers for events, pays for access for a defined period and receives those events through a webhook.
The concept can extend far beyond one product: machines purchasing tiny units of digital infrastructure from other machines as needed.
Blockchain also creates a natural way to limit an agent’s spending. Rather than handing autonomous software an unlimited account login, a user could give it a defined on-chain allowance and let the agent operate within that boundary.
Trouw believes AI may therefore become the environment in which a genuine token economy finally makes intuitive sense. Humans bundle services because that is how we think. Machines can evaluate, authenticate and pay for them one tiny unit at a time.
05 · What Comes Next Building With AI, and Building for AI
XYO’s AI strategy now runs in two directions. Internally, the company is using AI to accelerate software development and build products faster. Externally, it is designing products in which AI itself becomes a component of the application, while XYO Layer One supplies provenance, hashes and cryptographic verification around those systems.
Crypto Cards is one early demonstration of that model. Robotics, agent infrastructure and machine payments point toward the larger opportunity.
The significance is less about proving that every AI application belongs on a blockchain. Trouw explicitly argues the opposite. Most combinations of fashionable technologies become useful only when there is a real problem underneath them.
Autonomous AI is beginning to create that problem.
Software can now take actions, use private information, interact with other systems and eventually spend money without a person supervising every decision.
Trusting the operator’s promise may work much of the time. At machine scale, however, the more interesting question is whether the system can produce evidence that survives even when that trust disappears.
The most important AI guardrail may not be another policy explaining what a machine shouldn’t do. It may be infrastructure that makes the record of what it did impossible to quietly rewrite.
This feature is based on an exclusive interview conducted by Crypto Coin Show with Arie Trouw, Co-Founder, CEO and CTO of XYO, on 31 August 2026. The conversation covered AI sovereignty, cryptographic provenance, autonomous agents, robotics, machine payments and XYO Layer One.
