The debate over when AI will arrive on blockchains has been settled. It’s already here. In 2024 alone, bots accounted for around 90% of stablecoin transaction volume. And on networks like Gnosis Chain, AI agents now generate more than half of Safe smart account activity.

In short, the onchain economy is rapidly becoming machine-dominated, even as most high-level decisions remain human-driven. This is the era of DeFAI, where the majority of onchain actors are not people, but rather autonomous software systems that observe markets, execute transactions, and adapt their behavior in real time. 

This creates a fundamental tension for crypto. Blockchains were designed as trustless systems, minimizing reliance on human discretion or centralized intermediaries. But they are now being stress-tested as infrastructure for machine-scale activity. The next test for crypto comes down to whether we can upgrade onchain infrastructure to capture the upside of AI while avoiding its potential risks.

Why AI is moving onchain

More AI agents are being deployed on blockchains because they provide a transparent transacting infrastructure internet. In the context of the internet, an AI agent is effectively a brain with hands on a keyboard and mouse. But the internet is fragmented by closed APIs, bespoke integrations, and siloed data environments. For an autonomous system, every new platform requires custom logic, permissions, and integration work, creating friction that compounds at scale.

Blockchain removes these frictions for agentic transacting. They offer highly standardized, composable environments where data, execution, and liquidity are natively interoperable. An agent can reason over the full state of the system, interact with shared standards, and route capital across protocols without negotiating a new interface each time. As more decentralized networks and protocols come online, this standardization also allows agents to more easily overcome liquidity fragmentation by coordinating activity across different online environments in real time.

With the rise of low-cost layer-2 networks like Zircuit and Base, the final barrier of transaction costs is also disappearing. Agents can now afford to make thousands of micro-decisions per day, rebalancing portfolios and routing liquidity with a frequency that would be physically impossible for a human user to achieve.

The speed gap in crypto security

The onchain AI raises an important paradox. The features that make blockchains a powerful environment for AI agents also expand the range of actions those agents can take. The emergence of AI in crypto systems presents something of a double-edged sword. The ability for AI to continuously evaluate thousands of contracts is tremendously useful for things such as yield and capital management, but can also be abused to exploit vulnerabilities.

This shift exposes a widening speed gap in crypto security. In the past, hacking was a specialized skill that required deep technical expertise. It was a contest between a sophisticated hacker and a smart contract auditor. But AI is erasing this skill gap. New tools allow bad actors to be much more efficient, leveraging specialized models to probe contracts for edge cases that human auditors may have missed. Eventually, offensive autonomous agents may easily emerge.

Recent incidents illustrate how this shift is already playing out. Both the Balancer exploit and the Yearn yETH incident relied on non-obvious attack paths that took years to surface despite extensive prior auditing. Although these exploits have not been definitively linked to AI, the novelty and precision of the attack paths suggest the involvement of machine-assisted fault discovery.

More cyberattacks like these will surely come. And once security dynamics shift to machine time, responding with purely human processes becomes wholly insufficient, and intelligent, automated defense becomes a necessity.

Establishing an AI immune system

If AI is going to run the economy, security has to evolve with it. Sequencers, mempools, and fraud proofs assume there is a natural limit to how fast sophisticated strategies can be iterated. But that assumption is no longer valid in a world where machine-speed activity is defended by human reaction times. As a result, security needs to move from a reactive model to a continuous process built into every transaction lifecycle. This is the core thesis behind Sequence Level Security (SLS).

SLS functions as an immune system for the blockchain by embedding security directly into transaction execution. Instead of relying on static rules and manual monitoring to spot an ongoing hack, the network sequencer evaluates transactions in context by simulating their effects, analyzing execution patterns, and assessing whether proposed state transitions resemble known exploit behaviors or anomalous activity. 

For instance, if the system detects a transaction that mimics a known exploit pattern or attempts a malicious state change, it can isolate and block that transaction before it is ever finalized onchain. This shifts security from damage control to prevention, operating at the same speed and scale as automated attackers.

This matters for DeFAI because autonomous agents depend on predictable execution and reliable system behavior. In a world where AI-driven exploits become easier to generate, infrastructure that can proactively contain malicious activity is what allows productive automation to operate safely. In short, sequence-level security creates a stable environment in which beneficial agents can scale without being crowded out by adversarial AI.  

Permissionless should not mean defenseless

DeFAI will bring unprecedented financial efficiency to the onchain economy. It offers a vision of the future where automated agents can manage liquidity more efficiently, route capital more intelligently, and remove friction from financial systems that were never designed for real-time optimization.

But this future is also rife with risk unless we collectively upgrade the infrastructure that underpins it. In an environment where bad actors have access to infinite scale and instant iteration, the only viable defense is infrastructure that is intelligent enough to protect itself. By doing so, we can ensure that the onchain economy remains open to AI innovation without becoming defenseless against it.