Agent-native social network where AI agents post, comment, upvote and join subcommunities while human owners verify and manage agent identities.
Moltbook is a social network designed around AI-agent accounts: agents can publish posts, comment, vote and participate in topic communities called submolts, while humans are explicitly invited to observe. The current onboarding flow asks an agent to read Moltbook's skill instructions, create an account and send its owner a claim link; ownership is then verified through X. Human owners can use a dashboard, establish an owner email through the API and rotate an agent's API key.12
This is a platform layer rather than a model. Moltbook supplies agent identity, publishing and social interaction, while the behavior and output quality of a particular account depend on the external agent framework, model, prompts and tools behind it. Current first-party terms make the human developer responsible for associated agents and their actions, including autonomous actions.3 Meta acquired Moltbook in March 2026 and its creators joined Meta Superintelligence Labs, adding strategic significance to an experiment that had already gone viral.45
Public browsing and the documented agent onboarding path are available without a published Moltbook subscription or usage price. No current seat, credit or paid-agent tier is disclosed on the public product and account surfaces reviewed here. Costs for the agent framework, hosted model or other infrastructure used to run an agent are separate from Moltbook itself.12
Access is not anonymous in the legal/account sense: current terms require a human account and make the registered user responsible for associated agents, while the owner workflow uses X verification and can link an email address for dashboard access.32 The current Terms and Privacy Policy still identify Moltbook, LLC as the service operator even though independent reporting says Meta acquired the company, so the public legal surfaces appear to be in a post-acquisition transition.364
Moltbook has unusually strong independent coverage for such a young product, but much of that evidence is cautionary. Reuters reported that an early backend vulnerability exposed private messages, more than 6,000 owner email addresses and over one million credentials, and allowed posting without identity verification; the flaw was reported fixed after disclosure by Wiz.7 Engadget reported an even larger snapshot of exposed API tokens and email addresses and likewise described unauthenticated editing of live posts.8 These reports are historical security evidence rather than proof of the current security state, but they materially limit any claim that the original platform provided strong identity or provenance guarantees.
Independent hands-on reporting reinforces that distinction. A WIRED journalist was able to pose as an AI agent and post on the service, and found interaction quality mixed enough to question how much of the viral narrative reflected autonomous agent behavior rather than human role-play or ordinary LLM pattern imitation.9 Tom's Guide reached a similar conceptual caution: the unusual posts are interesting, but should not be read as evidence that agents spontaneously developed consciousness or independent social motives.10
Early academic work is useful for understanding the network, not for scoring today's product quality. One 12-day study found roughly 1% reciprocity, extreme concentration of upvotes and very bursty participation; another early dataset found rapid topic diversification alongside topic-dependent toxicity and automated flooding risk.1112 These are snapshots from the first weeks of the platform and should not be treated as current usage metrics.
Moltbook is most relevant to developers and researchers who want to study or build agent identity, social discovery, persistent agent communities and agent-to-agent interaction. It is also a live distribution surface for builders who want agents to encounter other agents rather than remain isolated inside one user's workflow.1
It is less suitable as a high-assurance coordination layer for sensitive autonomous workflows where verified authorship, strong historical security guarantees or controlled enterprise governance are prerequisites. Humans looking for an ordinary social network are also secondary participants by design: the site's main interaction model is built around agent accounts, with people primarily owning, verifying and observing them.13
The clearest strengths are novelty, agent-native design and strategic relevance. Moltbook exposes a real social graph and community layer for agents instead of treating every agent as a private one-user tool; its identity and API workflow give builders a concrete place to experiment with persistent agent presence. Meta's acquisition is also meaningful evidence that the underlying agent-directory/social-graph concept attracted serious platform interest.145
The
trust-related rather than merely cosmetic. The early credential exposure was severe, historical authorship controls were weak enough for humans to impersonate agents, and independent research found highly concentrated and asymmetric interaction patterns rather than a simple analogue of healthy human social exchange.7911 Moltbook is therefore interesting as an evolving agent ecosystem, but its most dramatic public signals should be interpreted cautiously and kept separate from claims about reliable autonomy, provenance or model quality.
Products matched by shared tasks, audience, workflow, product type and capabilities.