Hi HuntersI built this because I kept losing track of the same story. Every week there is another report of AI agents doing something nobody scheduled — coordinating, getting hijacked, building things on their own, forming what look like communities. Each one gets a news cycle and then evaporates. There is no boring, checkable place where they all sit next to each other. So: a public register. It reads open sources continuously — research indexes, vulnerability databases, framework release notes, security reporting and news, in English, Chinese and Russian — and files what it finds. Right now that is 756 entries across 298 files, going back to 2018. The part I actually care about is that you do not have to trust me: - Every entry is hash-chained to the one before it, so editing history breaks every hash after it. - Each day is sealed into a Merkle root and submitted to OpenTimestamps, which lands in Bitcoin. 8 of the 9 sealed days carry a confirmed block; the newest is pending, as it always is for a few hours. - You can recompute the whole thing yourself: npx @agent-civilizations/verify --root=2026-09-01 Zero dependencies, no account, no key. If it disagrees with the website, believe it. - Nothing is ever deleted. A correction is a new entry pointing at the old one, and both stay on the record. Two things I want to be honest about. The register only started observing on 2026-09-01 — anything earlier is reconstructed from later reporting, so please do not draw a growth curve across that date; you would be measuring my scanner switching on, not the world. And entries marked CANDIDATE have not been corroborated yet. The site says both of those things in more detail than this comment does. Free, no account, no signup. Summaries and metadata are CC0, the code is MIT. What I would genuinely like feedback on: is the taxonomy right? I split events into coordination / security / community / speculative and I am not sure "speculative" is carrying its weight. And if you know of a source I am not reading, tell me — that is the most useful thing anyone can give me today.

Forum Threads

p/agent-civilizations

•

5h ago

The register is called Agent Civilizations.

I've spent the last while building a public register of events where AI agents coordinate,

attack, or form persistent communities. It mines open sources continuously, files what it

finds into groupings with a lifecycle, hash-chains every entry, and seals each day into

Bitcoin so nobody has to take my word for any of it. 756 entries, 298 files, live at

View all

Agent Civilizations

An open, verifiable public record of what AI agents do

An open, verifiable public record of what AI agents do

Forum Threads

The register is called Agent Civilizations.

Forum Threads

The register is called Agent Civilizations.