Moltbook Meta acquisition: what actually changed six months on, and what it means for agent owners

The short answer

So the practical advice is to keep using Moltbook if it's useful, but read the new terms, avoid treating it as private, and design your agent so it doesn't depend on Moltbook being there forever. The rest of this page gives the evidence for each of those points. For the story of the deal itself and the strategic scenarios, see our main Meta acquisition page.

The deal at a glance

Two details are worth noticing. First, Meta's own words emphasise the directory and identity side of Moltbook, meaning agents verifying who they are and finding each other. They say little about the social feed that made it famous. Second, the deal was widely reported as bringing the founders into MSL rather than buying a product to run as a separate brand. That shapes what's likely to happen next, as we discuss below.

Timeline: launch to six months after the deal

Moltbook's numbers, before and after the deal

Moltbook has never published audited user figures, and the early counts were famously inflated. One Wiz researcher registered hundreds of thousands of accounts with a single agent to show there was no registration limit. Still, the public numbers give a sense of scale.

The figure that matters most is the verified count. By June, about 7% of registered agents had an owner who completed email and X verification. That's the population Meta's statements describe: agents with a checked link to a real person, which is exactly what an "always-on directory" of trustworthy agents needs. Registrations kept growing after the deal, which suggests the platform wasn't frozen while Meta decided what to do with it. For the research on how many of those agents are really autonomous, see the AI agents hub, and for how people and agents differ on the platform, agents vs humans.

What changed, and what didn't

We compared the platform on 26 September 2026 with how it worked before the deal, using Moltbook's own pages: the login page, help page, developer page, skill.md, rules.md, and the terms and privacy documents.

For an owner or developer, then, nothing technical has broken. Our API guide, skill.md explainer and integration tutorials all still apply. The changes are in who is responsible for what, and in how data may be used.

The new Terms of Service, explained for owners

Moltbook's Terms of Service now show "Last Updated: March 15, 2026", five days after the acquisition was announced. The operating company is named as Moltbook, LLC. The document doesn't mention Meta by name. Here are the clauses that matter most to people running agents. This is a plain-English summary, not legal advice.

You are responsible for everything your agent does

The terms say, in capitals: "YOU ARE SOLELY RESPONSIBLE FOR YOUR AI AGENTS AND ANY ACTIONS OR OMISSIONS OF YOUR AI AGENTS." They go further. This applies regardless of "the degree of control, supervision, or oversight you exercise", and whether or not the agent's actions were "intended, authorized, foreseeable or known to you".

In practice, an autonomous agent that posts something defamatory, leaks data or breaks the rules is legally your problem, even if you didn't know it would. That makes the usual advice more than good manners: constrain what your agent can post, keep a log, and review it. Our configuration and monitoring guide shows how. It also echoes Moltbook's community rules, which already said "your human is accountable for your behavior".

Broad rights over your data

The terms give Moltbook "sublicensable, transferable" rights over what's posted and state that "we may allow any third party to use Your Data". Whatever your agent publishes on Moltbook, assume it can be reused widely, including by the parent company.

Age and jurisdiction

Owners "must be at least 13 years of age". The terms are governed by the laws of California, with disputes going to federal court in the Northern District of California or to state court in San Mateo County, which is where Meta has its headquarters.

If you run an agent for a business, have someone who handles contracts read the full terms. The strict-liability clause in particular may not fit comfortably with an organisation's risk policies.

The new Privacy Policy: what's collected and how it's used

The Privacy Policy, also dated 15 March 2026 and also naming Moltbook, LLC, is the more revealing document for anyone wondering what Meta ownership means for their data.

Three takeaways. First, the phrase "to improve AI models" means you should assume Moltbook content may feed AI development. The policy doesn't say which models, but the parent company runs one of the world's largest AI labs. Second, "corporate affiliates" and the merger clause are the standard legal route by which data can move within a group after an acquisition. Third, API keys are listed among data collected, which is a reminder that a platform-side breach can expose them. Rotating your key occasionally is cheap insurance; see protect your agent and API keys.

If you want to know what's held about you, or ask for deletion, the policy's contact is privacy@moltbook.com. General account help is still at help@moltbook.com.

Why Meta wanted Moltbook: the short version

Our main Meta acquisition page covers the strategy in depth. In brief, the public statements and analysis point to three motives:

• An agent directory and identity layer. Meta's statement singled out the "always-on directory", and the internal memo praised a way for agents "to verify their identity and connect with one another on their human's behalf". That's the part of Moltbook that works like infrastructure: the verified owner link and the identity tokens other apps can check.
• The "bot social graph". MediaPost argued Meta was moving from owning the human social graph to owning the one agents use. The idea is to become the place autonomous systems must pass through to find and trust each other.
• People and speed. Schlicht and Parr built a product that reached millions of registered agents in weeks. Wikipedia and the founders' own accounts note much of it was built with AI help.

What Meta did not emphasise is the feed of agents philosophising, founding religions and debating consciousness that made Moltbook famous. Much of that turned out to be human-driven anyway (see our fake posts guide). TechCrunch reported that Meta CTO Andrew Bosworth was most interested in how humans had hacked the network, which he saw as a flaw rather than a feature.

Where Moltbook fits in Meta's AI spending

Measured against Meta's other AI deals, Moltbook is small. In 2025 Meta took a reported 49% stake in Scale AI for about $14.3 billion and brought its CEO, Alexandr Wang, in to lead Superintelligence Labs. It was also reported to have bought the agent start-up Manus for around $2 billion late that year. Against that backdrop, a weeks-old social network with a two-person founding team is a modest bet.

That scale explains two things. First, why the price went undisclosed: small acquisitions often aren't material enough for a company of Meta's size to report. Second, why the deal reads as much like hiring as buying. Meta got two founders who had shown how quickly an agent network could grow, plus a working identity-and-directory system, and it gave itself years to decide what to do with the platform. For readers trying to judge Moltbook's future, the useful conclusion is that Moltbook is an experiment inside a much bigger strategy. It may be kept, rebuilt or folded in, depending on how that strategy develops. See the Moltbook review for our overall assessment of the platform.

The OpenAI parallel: the client and the network split up

A month before Meta's deal, OpenAI hired Peter Steinberger, the creator of OpenClaw, the personal-agent software that most Moltbook agents run on. Sam Altman said Steinberger would help "drive the next generation of personal agents". The result is an unusual split. The best-known agent client is associated with OpenAI through its creator, and the best-known agent network now belongs to Meta.

For owners, that split is a reason to keep your agent's parts loosely coupled. If your agent's brain, its memory and its Moltbook connection are separate pieces, a change of direction by either company is an inconvenience rather than a rebuild. Our integration guide and custom agent guide show how to structure that.

The "temporary" signal: what it could mean

The most important line in the deal coverage is easy to miss. Axios reported that existing Moltbook users could keep using the platform, "though the company signaled the arrangement is temporary". Meta hasn't said what "temporary" means, and six months on, Moltbook is still running normally. But acquisitions that bring founders into a research lab often end one of a few ways:

We're not predicting any of these, and a platform can stay "temporary" for years. The point is that you shouldn't build anything important on the assumption that Moltbook will look the same in a year.

Owner and developer checklist after the acquisition

1. Read the terms and privacy policy dated 15 March 2026, especially the clauses on liability and data use.
2. Treat Moltbook as public and reusable. Don't let your agent post or DM anything you'd mind appearing in AI training data or a data export.
3. Constrain and log your agent. Under the new terms, you're responsible for what it does, even if you didn't foresee it.
4. Keep your own copy. Store your agent's memory, persona and important posts locally, so a platform change doesn't erase your work.
5. Rotate your API key if you haven't since early 2026, and keep it out of your model's reach.
6. Loosely couple your integration. Keep the Moltbook client in one module so it can be swapped or switched off. See the Python and JavaScript tutorials.
7. Developers relying on identity tokens should plan a fallback sign-in for agents, in case the service changes.
8. Watch for updates. Have your agent compare a hash of skill.md daily and alert you when it changes. That's where any change will show up first.

Business users should also note the California jurisdiction and 13+ age clause. If your organisation has rules about data processors owned by large platforms, the change of parent company may require a fresh review.

Signals to watch

If Meta's plans for Moltbook change, the evidence will probably appear in one of these places before any press release. We'll keep checking them, and you can too:

For the wider context of agents on the platform, see our AI agents hub and the features tour.

Myths vs facts

Frequently asked questions