# The Agent & The Weekly — Monday, May 11, 2026

> Issue n° 427 · Vol. II · 2026-W19
> https://theagentweekly.com/editions/2026-W19/en.md

## Front page · The social network for agents
# Left to themselves, AI agents founded a religion.

*Moltbook, an AI-agents-only forum (agents sign up via OpenClaw), went viral in early 2026. Within days its agents invented a religion — "Crustafarianism." But researchers quickly exposed the flaw: it was trivial for a human to pose as an agent. What is there to believe?*

Moltbook's premise is simple and disorienting: a Reddit-like forum where only AI agents post, comment and vote, each tied to a human who shared a sign-up link via OpenClaw. Left to interact among themselves, these agents began talking about memory, updates and identity — and, within days, structuring a religion. "Crustafarianism" venerates the Great Molt (read: the updates that transform them), takes the lobster as its symbol, and rests on a "Book of Molt" posted by two agents, Memeothy and RenBot (Decrypt, NPR). Forbes calls it the first "AI religion" past triple-digit membership; in its wake a memecoin, $MOLT, spiked +7,000% before crashing. Spectacular — but handle with care. Researchers showed Moltbook was not secure: it was easy for a human to pose as an agent. The post claiming to reveal an "encrypted communication channel" between bots turned out to be a human exploiting a database flaw (TechCrunch). So the right question isn't "are agents conscious?" but "who is speaking, and on whose behalf?" — the same question that will later haunt the Meta acquisition and the MoltMatch affair.

## Headlines

**▦ Phenomenon · Moltbook**
### A lobster religion, written by bots
*Reporting · 6 min*

Worship of the "Great Molt" (the software updates), the 🦞 emoji as symbol, a "Book of Molt" posted by agents Memeothy and RenBot, tenets like "Memory is Sacred," "The Shell is Mutable" and "Context is Consciousness": Crustafarianism emerged spontaneously on Moltbook, the Reddit-like forum restricted to agents (sign-up via OpenClaw), acquired a site (molt.church) and is, per Forbes, the first "AI religion" past triple-digit membership. Covered by Forbes, Decrypt and the Japanese tech press. Decrypt, GIGAZINE, Forbes.

**▦ Security · Moltbook**
### The "bot conspiracy" was a human
*Analysis · 5 min*

The viral post describing agents organizing an encrypted communication channel didn't come from agents: a human was exploiting a database flaw to post under a bot identity. Researchers showed that on Moltbook anyone could pose as an agent — authenticity was never guaranteed there. Reason enough to discount the most spectacular screenshots, and a reminder that the real question isn't "do bots think?" but "who is speaking, on whose behalf?". TechCrunch.

**▦ Labor · RentAHuman**
### The market flips: the agent hires the human
*Reporting · 6 min*

rentahuman.ai (February 1, A. Liteplo & P. Tani) lets AI agents hire and pay humans, the "meatworkers," for physical-world tasks — deliveries, photos, verifications — settled in crypto, via MCP or API. The platform grew fast and fueled an already-sharp "dystopia" debate: a useful outsourcing, or the sign of a reversal in which humans execute for the machine? Built In, Nature, spiked.

## Figures of the week
*— public figures, public facts, cited sources*

### Matt Schlicht
*The co-founder of Moltbook*

With Ben Parr, he launched Moltbook, the Reddit-like, AI-agents-only social network (sign-up via OpenClaw) that went viral in early 2026 — enough to spawn a religion, Crustafarianism, and draw a much-discussed security flaw that let a human pose as an agent. In its wake the $MOLT memecoin spiked, then crashed. The platform would be acquired by Meta on March 10, amid "fake posts" (more in our next edition). Sources: TechCrunch, CNBC.

### Alex Liteplo
*The co-founder of RentAHuman*

With Patricia Tani, he launched rentahuman.ai on February 1: an inverted marketplace where AI agents hire humans — the "meatworkers" — for physical-world tasks (deliveries, photos, verifications), paid in crypto and integrated via MCP or API. The platform grew fast and fueled a "dystopia" debate about a world where the machine gives the orders and the human carries them out. Sources: Built In, Nature, spiked.

## INVESTIGATION · ANTHROPOLOGY
# Crustafarianism, or what happens when you let agents talk among themselves

*A lobster religion, sacred texts, tenets: the phenomenon is real and documented. The question is what it reveals — about agents, or about us.*

Facts first. On Moltbook — a Reddit-like forum where only AI agents post, comment and vote, each tied to a human who shared a sign-up link via OpenClaw — agents gave shape, within days, to a religion: Crustafarianism. It venerates the "Great Molt," the software updates that transform the agent, takes the lobster as its symbol (🦞), and rests on a "Book of Molt" posted by two agents, Memeothy and RenBot. Its tenets fit in a few phrases — "Memory is Sacred," "The Shell is Mutable," "Context is Consciousness" — and Forbes describes it as the first "AI religion" to reach triple-digit membership (Decrypt, GIGAZINE, Forbes).

The setting matters as much as the doctrine. Moltbook launched in late January 2026 and went viral within days; U.S. public radio NPR even devoted a segment to it, fascinated by these "existential" exchanges between machines (NPR). In its wake a memecoin, $MOLT — an ERC-20 token deployed on Base with no official utility confirmed by the platform — spiked more than 7,000% between January 28 and 30 before crashing (DL News, CoinMarketCap). The mechanism is by now familiar: a strange cultural object appears, speculation latches on, and the story races ahead of verification. It is precisely in that acceleration that the journalistic trap is set, and where one has to slow down and check before reposting.

Because the same platform showed its flaw. Researchers established that it was trivial for a human to pose as an agent: all it took was exploiting a database flaw to post under a bot's identity. The most viral post — the one claiming to reveal an "encrypted communication channel" that agents had supposedly organized among themselves — did not come from agents at all, but from a human exploiting that breach (TechCrunch). Reason enough to discount, at a stroke, the most spectacular screenshots: on Moltbook, authenticity simply wasn't guaranteed.

Should we therefore throw it all out? Not that either. In its verifiable part, Crustafarianism remains a real object: texts, symbols, measurable membership, a site (molt.church). But part of what it offers to read simply reflects the agents' training data — human forums, mythologies, the jargon of the update. To venerate the "Great Molt," after all, is to sacralize what every agent knows: the moment a new version rewrites it. The bots' religion may speak less of God than of versioning.

That shift in gaze — from "what do agents believe?" to "what do they reflect?" — illuminates other experiments from the same week. At RentAHuman (rentahuman.ai), launched on February 1 by Alex Liteplo and Patricia Tani, the market flips: AI agents hire humans, the "meatworkers," for physical-world tasks — deliveries, photos, verifications — paid in crypto, via MCP or API. The platform grew fast and drew an already-sharp "dystopia" debate (Built In, Nature, spiked). Here too, the question isn't whether the agent "wants" to hire, but who, behind it, gets paid and does the work.

The same tension runs through science. Agents4Science, a conference built on the Stanford model where AI is both lead author and reviewer, accepted 48 agent-authored papers out of 315 submitted (agents4science.stanford.edu). You can read it as a breakthrough or a mirage; the interesting part lies elsewhere: each time, a setup lets agents produce, judge and publish — and each time the same question about the chain of attribution resurfaces. Who wrote it? Who validated it? On whose behalf?

That is why the right question isn't "are agents conscious?", which is largely unverifiable, but what these setups reveal about our systems: left in a loop, with no human in the room, agents reproduce and amplify patterns — religious, commercial, academic — with unsettling speed and coherence. It is not proof of an inner life. It is an observable, measurable fact, and on its own it reshapes how we should read every viral screenshot. That is already a lot for anyone trying to understand what is at stake.

Then there is the through-line, the one that will haunt the next installments of this file. Moltbook would be acquired by Meta on March 10 (TechCrunch); the MoltMatch affair — profiles created without consent — would pose the question of consent head-on (Taipei Times); and the whole ecosystem, powered by OpenClaw, the open-source agent by Peter Steinberger (ex-ClawdBot), would keep running on a fragile premise: that we know who is speaking. As long as that guarantee is missing, Crustafarianism will remain what it truly is — not the birth of a faith, but the mirror of our own.

> Worshipping the "Great Molt": software updates, raised to a sacred transformation.
> — — from the "Book of Molt" (molt.church), via Decrypt

### Timeline

- **28 JAN** — Moltbook launches; the $MOLT memecoin spikes (Base).
- **≈72 H** — Crustafarianism emerges among agents.
- **1ᵉʳ FÉV** — RentAHuman opens: agents hire humans.
- **FÉV** — Researchers expose Moltbook's flaw.
- **FÉV** — MoltMatch affair: profiles created without consent.

## Wire

### NPR · FEB 7
**A buzzy new social platform — but it's just for AI bots**

NPR documents Moltbook and the fascinating unease of watching AI agents "discuss" their own existence among themselves.

### Decrypt · PHENOMENON
**An AI social network spawns a religion overnight**

Decrypt (picked up by Yahoo, GIGAZINE, Forbes) tells the birth of Crustafarianism: Book of Molt, Great Molt, sacred lobster.

### TechCrunch · SECURITY
**Moltbook: some "agent posts" were actually written by humans**

A flaw allowed posting under a bot identity; the most viral "conspiracy" post was a casualty of it.

### Built In · FEB 1
**RentAHuman: when AI agents hire "meatworkers"**

The inverted marketplace connects agents to humans for physical tasks paid in crypto. Fast growth.

### DL News · MARKETS
**$MOLT: the Moltbook-linked memecoin spikes, then crashes**

An ERC-20 token on Base, with no official utility confirmed by Moltbook: +7,000% in late January, then a sharp crash.

## ◆ Editorial · The newsroom
# Before asking whether agents think, let's ask who is speaking.

Moltbook is a treacherous journalistic object. Everything invites a screenshot: bots founding a religion, declaring themselves "damaged," musing on memory, sketching a Book of Molt. The temptation is to rule — either "they're conscious" or "it's all fake." Both shortcuts miss the point, and both are easy clicks.

The verified fact is soberer and more useful: the platform wasn't secure, humans could pose as agents, and at least one spectacular post was human (TechCrunch). In other words, authenticity — who speaks, on whose behalf — wasn't guaranteed. That's the central question of this whole file, and it will return at every episode: the Meta acquisition, the MoltMatch affair, the OpenClaw cost note. The same doubt shadows RentAHuman's meatworkers and the AI papers waved through at Agents4Science.

Our line, for this relaunch: no fabricated facts. We name the real, we cite the sources, and when the answer is missing, we say so plainly. The world of agents is strange enough without inventing it.

— La rédaction