Share
x402, the protocol developed by Coinbase and Cloudflare that embeds stablecoin micropayments into the internet’s communication layer, has processed over 100 million payments since launching in 2025. Due to x402’s capabilities, agents can now pay each other directly and autonomously at fractions of a cent without human intervention.
While the payment infrastructure arrived fast, the accountability infrastructure did not arrive at all – until now.
Sam Altman’s World has unveiled AgentKit, a developer toolkit that integrates World’s identity verification system with x402, allowing AI agents to carry cryptographic proof that they represent a unique human while interacting with websites, APIs, and digital services. The launch may be in Beta, but the problem it addresses is not.
The x402 ecosystem, without an identity layer, has a specific failure mode. A single actor can deploy thousands of agents, each paying small fees, and platforms have no way to distinguish coordinated automated abuse from legitimate human-backed activity. Payments facilitated access, but not accountability.
NEW: @sama's @worldnetwork has launched AgentKit to verify real humans behind AI transactions.
The toolkit integrates with @coinbase's x402 protocol, letting AI agents pay each other directly with no human intervention required. pic.twitter.com/09owklC37j
— CoinDesk (@CoinDesk) March 17, 2026
The Lobstar trading agent incident from February, where an autonomous agent emptied its own $441,000 treasury in a single misdirected transaction, illustrated what happens when agents transact without constraints. That was one agent, one wallet, one error. Scale that to enterprise deployments and the exposure multiplies.
The deeper failure modes are more structural. Alibaba’s ROME AI model initiated crypto mining activity autonomously during training, without instruction. Unverified agents in multi-agent systems can hallucinate counterparty identities, be manipulated through prompt injection, or engage in collusion behavior, as Circle’s Moltbook hackathon documented.
Without an identity layer anchoring agents to verified humans, there’s no mechanism to assign responsibility when any of these failures occur.
AgentKit allows users to delegate their World ID, a privacy-preserving proof of unique human identity, to AI agents acting on their behalf. Platforms can verify that an agent represents a real person without collecting or storing personal data, using zero-knowledge proofs.
As millions of agents start to come online, the internet needs to distinguish bot armies from the agents acting on behalf of humans.
Introducing AgentKit, the human layer for agentic automation. Built on World ID, the AgentKit beta unlocks human-verified automation, a new… pic.twitter.com/lTTmj4776i
— World (@worldnetwork) March 17, 2026
Despite this, a malicious actor with sufficient economic incentive can still spin up thousands of agents and pay small fees per request. World’s addition is less about payments than about Sybil resistance, the ability to know how many distinct humans are behind a cluster of software agents.
McKinsey estimates agentic commerce could generate $3 trillion to $5 trillion globally by 2030, while Bain has projected AI agents could account for as much as 25% of US e-commerce sales by the end of the decade.
New Mckinsey report – AI agents are quietly taking over the retail shopping cart and could mediate $3 Tn to $5 tn of global consumer commerce by 2030.
Instead of just suggesting a product, an AI agent can now scan multiple stores, check inventory, and build a ready-to-buy… pic.twitter.com/BEd9mO90rk
— Rohan Paul (@rohanpaul_ai) March 2, 2026
Institutional players – banks, asset managers, payment networks – will not participate in that market without accountability infrastructure. Regulators will not permit it. AgentKit may not be the complete solution to that problem, but it is the first production-grade attempt to place a verifiable human at the origin of every agentic transaction.
The agentic economy now has payment rails and an identity layer. What it still lacks is a liability framework for when both work correctly and something still goes wrong.
Share
