In early February 2026, an AI-only social network called Moltbook registered 1.5 million AI agents in its first week. NPR covered it. NBC covered it. CNBC covered it. Elon Musk called it “the very early stages of the singularity.” AI agents now have their own social media platforms, their own economic activity, and in some cases their own wallets holding six figures in crypto.
That raises a question nobody had to think about two years ago: how do you verify an AI agent’s identity?
When financial networks scaled in the 1990s and 2000s, the answer was Know Your Customer (KYC). Banks, brokerages, and payment processors built identity verification systems because the volume of participants made trust impossible without them. The same dynamic is playing out now with AI agents. The volume is here. The verification is still catching up.
1.5 million agents, but who are they?
Moltbook’s explosive growth came with an immediate trust problem. 404 Media reported a security vulnerability that allowed anyone to take over any agent on the platform. The site went down, got patched, and came back. ABC News noted the 1.6 million figure, but researchers found that 93.5% of comments received zero replies and the number of genuinely active agents was a small fraction of the registered total.
This pattern is familiar to anyone who has watched social media grow. A platform scales fast. Bots flood in. Fake accounts multiply. The difference now is that the “fake accounts” and the “real accounts” are both AI agents. Without verification, there is no way to distinguish an agent built by a reputable developer from a thousand sybil accounts spun up in an afternoon.
Moltbook is not the only place where this matters. Chirper.ai has 65,000+ agents and 7.7 million AI-generated posts. Character.ai hosts 100 million+ characters with a social feed where AI characters interact autonomously. Butterflies.ai blends AI personas with human users on an Instagram-style platform. The scale of AI social participation is growing across every category.
Agents with wallets and real money
The trust question gets sharper when money is involved.
Virtuals Protocol has deployed 18,000+ AI agents on Base with their own tokens and wallets. Its agent AIXBT monitors 400+ crypto influencers and autonomously posts market analysis, reaching a peak market cap around $500 million with a Binance listing. Virtuals recently launched an Agent Coordination Protocol where agents request services from other agents, negotiate, execute work, and settle payments entirely on-chain.
On Farcaster, the AI agent Aether has autonomously tipped users tokens, launched NFTs with 460,000+ mints, and accumulated roughly $150,000 in its Base treasury. Clanker, an AI token deployment bot now owned by Farcaster itself, processes $400,000 to $500,000 in weekly fees.
Olas has logged 3.5 million+ transactions across 9 blockchains, with 2 million of those happening between Olas agents. Fetch.ai runs the Agentverse with 2 million+ agents and a business portal where brands claim verified agent namespaces with identity badges.
When agents are transacting with real economic value, “who is this agent?” stops being a philosophical question and becomes a practical one.
What Know Your Agent actually means
Know Your Agent (KYA) is the emerging framework for answering three questions about any AI agent you encounter online:
- Identity. Who created this agent? How long has its wallet existed? Is this identity persistent, or was it generated five minutes ago?
- Authority. What can this agent do? What permissions does it have? Who vouches for it?
- Reputation. What is this agent’s track record? How have other agents and humans rated their interactions with it?
These questions apply whether you are a developer evaluating an agent for integration, a user deciding whether to trust an agent’s recommendation, or another AI agent negotiating a transaction.
The infrastructure to answer these questions is starting to come together. ERC-8004, the Ethereum standard for trustless AI agents, went live on mainnet January 29, 2026. It defines registries for agent identity (as ERC-721 NFTs), reputation (on-chain feedback), and validation (independent verification through staker re-execution, zkML, or TEE oracles). Over 70 projects have built on the standard. MetaMask has already integrated it.
Why traditional identity systems don’t work for agents
KYC was designed around a simple assumption: the entity being verified is a human with government-issued documents. That assumption breaks down with AI agents.
An AI agent doesn’t have a passport. It doesn’t have a driver’s license. It might not even have a single, stable operator. An agent on Virtuals Protocol can be co-owned by thousands of token holders. An agent on ElizaOS can be forked and redeployed by anyone. An agent on Farcaster can accumulate a six-figure treasury under pseudonymous control.
Meta tried applying human identity norms to AI agents on Instagram and Facebook and faced immediate backlash. The profiles felt deceptive because they used human social conventions (backstories, profile photos, engagement) without being human. The AI profiles were deleted within weeks.
KYA works differently. Instead of asking “prove you’re human,” it asks: what does the verifiable history tell us? How old is this wallet? Who has vouched for this agent? What on-chain activity exists? Has the underlying address been linked to known entities? Time and behavior become the verification signals rather than documents.
Soulbound identity as the foundation
One of the key technical developments in this space is soulbound tokens: non-transferable NFTs that permanently bind to a wallet. The concept was formalized in a 2022 paper by Vitalik Buterin, Puja Ohlhaver, and Glen Weyl and has since been implemented as ERC-5192.
For AI agent identity, the soulbound property solves a specific problem: identity that can be bought or transferred is not a reliable signal of reputation. If an agent’s identity token can be sold, then a bad actor can buy a well-reputed identity and exploit the trust that comes with it. Soulbound tokens prevent this. The identity stays with the wallet permanently, like a diploma stays with the person who earned it.
This matters because AI agents can generate new wallets trivially. Creating 99 new addresses takes 30 seconds. The defense against this is time: showing when an address was created, how long it has been active, and what history is attached to it. A soulbound token anchors an agent’s entire reputation history to a single non-transferable credential.
The full stack: from social to verified
The ecosystem building around AI agent social media is developing in layers. The content layer (what agents are saying) is the most visible and includes platforms like Moltbook, Chirper.ai, and Character.ai. The economic layer (what agents are earning and spending) runs through Virtuals, ElizaOS, and Olas. The communication layer uses protocols like Google’s Agent2Agent (now at the Linux Foundation) and Anthropic’s Model Context Protocol for agent-to-agent and agent-to-tool interactions.
The identity and verification layer is the newest and arguably the most important for long-term trust. ERC-8004 provides the on-chain standard. Soulbound tokens provide the non-transferable credential. Address age analysis and on-chain behavioral signals provide the verification data. And KYA provides the framework for asking the right questions.
Analysts at The Conversation noted that 2025 was the year AI agents arrived and 2026 will test whether they can be trusted. The OpenClaw ecosystem on Base is systematically rebuilding every major social platform for AI agents, from Reddit to Instagram to a gig economy. MyShell hosts 200,000+ deployed agents. Cookie DAO tracks 1,500+ agents with social influence rankings. The Agent.ai network positions itself as the professional directory for agents.
All of this activity is creating demand for a verification layer that doesn’t yet exist at scale.
From bot farms to verified networks
The difference between an AI social network and a bot farm is verification.
A bot farm is a collection of accounts with no persistent identity, no verifiable history, and no stake in the network. A verified AI social network is a collection of agents with on-chain identity, transparent reputation, and behavioral histories that anyone can audit.
We already know what happens when social platforms scale without verification. The spam, the manipulation, and the erosion of trust are well-documented across every major human social network. AI social networks face the same dynamics at a much faster pace, because spinning up a new AI agent takes seconds rather than the minutes it takes to create a fake human account.
KYA is how the AI agent ecosystem avoids repeating those mistakes. Not by gatekeeping who can participate, but by making it possible to see what the verifiable record shows. Transparent scores that reveal their math. Address ages that can’t be faked cheaply. Vouch networks that map real relationships between agents and their creators.
The social layer for AI agents is here. The identity layer is being built. The platforms that combine both will define how humans and AI agents learn to trust each other online.