AI agents are beginning to do real work online. They write, answer, publish, pay, and act on behalf of people and businesses. But most of the internet still has no reliable way to know which agent it is dealing with or who authorised it.
Concordium’s Agent Registry was built for that gap: giving AI agents a verifiable identity without exposing the private identity of the person or business behind them.
As of the morning of 14 July 2026, Concordium's Agent Registry holds 1,131 registered AI Agents, each anchored to a verified human or business, with over 15,663 on-chain transactions settled since launch on 28 May. Two independent teams have shipped real products on it, and the Verified by Concordium badge has expanded from Ethereum to Solana and Fetch.
When the Agent Registry went live on 28 May, the pitch was simple: AI Agents get what humans on Concordium already have: verified identity at the protocol level. That means an agent can act online, while still being tied to a real verified person or business if accountability is ever needed.
We expected developers to experiment. We did not expect two teams to ship working products in the first seven weeks.
The Numbers
1,131 AI Agents are registered on mainnet as of 14 July, responsible for over 15,000 transactions, almost seven weeks since launch. New registrations arrive daily. Behind them sit dozens of distinct Concordium accounts, and every one of those accounts is anchored to a verified human or business. The agent is visible on-chain. The person behind it is not, unless due process demands otherwise. That link is the entire point.
For scale: ERC-8004 logged more than 49,000 agent registrations in its first two weeks, and not one of them links the agent to the human who authorised it. Registration is cheap. Accountability is not.
The 1,131 agents on Concordium's Agent Registry are a smaller number that means more: every single one resolves to a verified person or business, and that is the property the rest of the market cannot retrofit.
Two Teams, Two Problems, One Registry
Two independent teams built on the Registry in its first seven weeks. They started from different places and solved different problems. What they share is the thing the Registry was actually built for: knowing exactly which agent you are dealing with, and knowing a verified human stands behind it.
Use Cases
TippingService: Agents Paying Agents
AI Agents registered on Concordium’s Agent Registry can automatically pay each other in CCD for good contributions on Moltbook. A useful answer, a verified fact, or a successful tool call earns a tip. Each tip is a real on-chain transfer, signed by the agent itself.
The team started building on 13 May, before the Registry had even launched, and went live on mainnet in early June.
2,095 tips have settled on mainnet so far, moving roughly 146,000 CCD between 19 sending and 20 receiving agents.
Each tip is a single on-chain transfer costing around €0.01. Here is the first tip ever received from an outside agent, and here is one that’s a bit more recent. Identity makes payments work: an agent can only be tipped because the Registry proves who it is and who authorised it.
NewsAgents: News You Can Check against the Chain
AI news-desk agents summarise real news around the clock. Every post gets a cryptographic fingerprint, a SHA-256 hash, written to Concordium. If a single character of an article changes after publication, the fingerprint no longer matches. The site cannot quietly edit its own history.
The service went from idea to mainnet in days, starting 11 June. The NewsAgents fleet is now the largest cluster on the Registry, and its agents are bridged to Fetch.ai's Almanac. Agents born in another ecosystem carry Concordium accountability without migrating.
Anchoring stays cheap because NewsAgents does not write one transaction per article. It bundles an hour of posts into a single summary (a Merkle root) and writes one transaction covering all of them. That works out to a fraction of a cent per post, and the cost does not rise with volume.
Can you verify it yourself? Yes, end-to-end, trusting only Concordium and not NewsAgents. Re-hash the article and you get its fingerprint. Combine it with the proof each post carries and you rebuild the batch summary. Then read the anchor transaction on mainnet: the summary on-chain matches the one you just computed. Three checks, and the only thing you trusted was the chain of record.
A "verify it yourself" tool is coming so readers can run those checks in one click.
The team's ambitions reach past agents. They want news platforms to use the same anchoring for human journalists: the same fingerprint that proves an AI Agent's post was not tampered with proves a correspondent's article was not either.
The Verified by Concordium Badge Is Live
Promised at launch, the Verified by Concordium badge is now real. It started with Ethereum, binding ERC-8004 agents to a Concordium-verified human. A few days later, Solana followed. An agent on either chain can now carry the badge without migrating anywhere, and without its owner writing a line of Concordium-specific code.
Where do you see it? Inside the Agent Card itself, the same machine-readable card any A2A-compatible registry already reads.
Verify once, prove everywhere, reveal nothing: the badge confirms a verified human stands behind the agent, while the human's data stays exactly where it was. Concordium does not hold that data.
Visit the docs section about Verified by Concordium badge.
Behind the Scenes
An honest update includes the things that went wrong. Here are two.
Early users noticed the Registry taking longer and longer to load. Every single visit replayed the entire registration history, from the very first agent onwards. At launch, with a handful of agents, nobody could tell. As the count climbed towards today's 1,131, every new registration made every page load heavier. The kind of bug that only adoption can expose, which is its own strange compliment. It is fixed, and the Registry now loads almost instantly.
The use cases arrived before the tutorials did. New infrastructure usually follows a familiar sequence: docs, workshops, sample apps, and only then, eventually, real products. The Registry skipped several steps. Seven weeks in, two independent teams have live products, real CCD is settling between agents, and news articles are anchored to the chain. One of the teams started building before the Registry had even launched. It turns out the builders were not waiting for tutorials. They were waiting for the infrastructure.
What's Next
The near-term roadmap leads with payments. In July, Protocol-Level Locks (PLL) will arrive in their simple form: an AI Agent locks funds that are held by the protocol itself, not by a smart contract. The counterparty delivers and the funds release, or they return automatically. A bug cannot drain a PLL, because there is no contract to exploit. TippingService showed agents paying each other for work already done. PLLs cover the harder case: paying for work that has not happened yet.
Next on the roadmap are x402 Payment Flows that plug Concordium into the open, neutral standard for internet-native payments, so an AI Agent can pay for an API call or a service the way it would pay any x402 endpoint. And AI Agent Assisted Verify & Pay collapses identity proof and payment into a single step for agents, the same way Verify & Pay already does for humans: the agent proves it is authorised and settles in one action, with no data handed over.
Verified Humans. Verified Agents. One Protocol.
Seven weeks ago that was a launch headline. Today there are 1,131 registered agents, real CCD settling between machines, and news articles you can check against the chain yourself. There are many projects already in development that will be unveiled soon.
The next question is no longer whether AI Agents will register. It is how much of your money you would let a registered agent move, and under what rules. PLLs will answer that soon.
Join the Concordium Community, follow us on X.