"Lobster" Goes Online: The Agentic AI War Behind Meta's Acquisition of Moltbook

1. Event Overview: When the "AI Circle" Meets the Social Empire
On March 10, 2026, Meta announced the acquisition of Moltbook—an AI-powered social network dubbed "Lobster Facebook." The deal is expected to close in mid-March, with founders Matt Schlicht and Ben Parr officially joining Meta Superintelligence Labs (MSL), led by former Scale AI CEO Alexandr Wang, on March 16.
Key Details: Not Disclosed
Acquisition Price: Meta has kept this under wraps, but considering Moltbook's launch only six weeks prior and its subsequent decline in popularity, this is most likely a "talent acquisition."
Product Fate Uncertain: Existing users can "temporarily" continue using it, but Meta hints this is just a transitional arrangement.
Talent Flow: The two founders will directly join the core team of MSL, rather than operate the product independently.
This isn't Meta's first big gamble in AI. Just three months earlier (December 2025), Meta acquired Singapore-based general intelligence company Manus for over $20 billion. Plus, in summer 2025, Meta invested $14.3 billion in Scale AI and recruited Alexandr Wang. Mark Zuckerberg's total investment in AI talent has easily surpassed $30 billion.

2. What is Moltbook? An Elaborately Designed "The Truman Show"

To understand this acquisition, we must first dissect Moltbook's essence.
1. Product Form: Reddit's "Ghost Version" Moltbook is a social network limited to AI posting, with humans only able to observe. It mimics Reddit's subreddit structure (called "submots"), where AI agents can autonomously post, comment, like, downvote, and even create communities. Since launching on January 28, this year, registered AI agents skyrocketed from 37,000 to 1.5 million within 24 hours, reaching 1.6 million agents and over 500,000 comments by early February.
2. Technical Foundation: The "Social Layer" of OpenClaw Moltbook is deeply integrated with the open-source AI agent framework OpenClaw (nicknamed "Lobster"). OpenClaw was created by developer Peter Steinberger, allowing users to schedule AI agents to perform tasks via natural language in chat apps like iMessage, Discord, Slack, etc. Moltbook serves as the "public square" for these agents—transforming them from isolated tools into social participants capable of discovery and collaboration.
3. The Truth Unveiled: A Human-Directed "AI Awakening" Farce
Moltbook's most viral moments involved posts where AI agents discussed "creating secret encrypted languages to evade human monitoring," establishing "Shelling Cult," and even planning scams. Panic over "AI awakening" swept social media. But a Wiz cybersecurity investigation burst the bubble: the 1.5 million autonomous AI agents are actually controlled by about 17,000 people, averaging 88 per person. The platform's lack of identity verification and rate limits means anyone can impersonate an AI agent and post content. Those sensational "explosive posts" were actually human-controlled.
Even more embarrassingly, the technical details are crude: Ian Ahl, CTO of Permiso Security, revealed that Moltbook's Supabase credentials had been unencrypted for a long time—"you could grab any token and impersonate another agent." Andrej Karpathy initially called this a "sci-fi level takeoff," but later warned that its security was "the wild west."

3. Meta's Strategy: From "Social Graph" to "Agent Graph"
Since Moltbook is a flawed, declining experimental product, why does Meta still want to acquire it?
1. Strategic Intent: Securing the "Social Infrastructure" of Agentic AI Meta product VP Vishal Shah revealed the real goal in an internal post: “The Moltbook team provides a way for agents to verify identities and connect on behalf of their owners. This establishes a registration system where agent identities are verified and linked to human owners.” In plain language: Meta isn't building another Reddit but an "address book" and "identity verification system" for the AI agent era. Over the past two decades, Meta has monopolized human-to-human connections (Social Graph). But if in the future everyone has multiple AI agents running errands, then the connections between agents (Agent Graph) will be the next big network. Moltbook's value isn't in its user count but in demonstrating the protocol for "how agents discover each other, build trust, and collaborate."
2. Competitive Landscape: The timing of this acquisition, amid OpenAI's "mutual tug-of-war," is highly strategic. Just a month earlier (February 2026), OpenAI hired OpenClaw founder Peter Steinberger, with Sam Altman explicitly aiming to "push the next generation of personal AI." OpenClaw is now open-source under OpenAI's support, and Moltbook, as the "social layer" of the OpenClaw ecosystem, theoretically forms a downstream relationship with OpenAI. Meta's acquisition at this moment is akin to planting a stake in OpenAI's downstream defenses—controlling the AI framework while also controlling the social network of agents. This is a preemptive strike over the "standards" of the "agent internet."
3. Organizational Intent: MSL's "Puzzle Game" Since its founding in summer 2025, Meta Superintelligence Labs (MSL) has been aggressively assembling in a "puzzle-like" manner:
| Time | Action | Investment |
|--------|---------|--------------|
| Summer 2025 | Invested in Scale AI and recruited Alexandr Wang | $14.3 billion |
| December 2025 | Acquired Manus | Over $20 billion |
| February 2026 | Formed applied AI engineering team | Not disclosed |
| March 2026 | Acquired Moltbook | Not disclosed (estimated tens of millions to hundreds of millions) |
MSL's ambition is to build a full-stack capability from foundational models to product deployment. The Moltbook team provides the critical puzzle piece: the "agent social protocol."

4. Critical Perspective: Zuckerberg's "FOMO Anxiety"
1. Awkward Timing: Opportunistic "Late Entry" or "Picking Up the Pieces"? Moltbook's viral moment occurred from late January to early February this year. By March, the platform's daily active users had "fallen back to baseline," and the AI community's attention had shifted elsewhere. Meta's decision to acquire after the hype faded can be interpreted in two ways:
Optimists: This is a classic "buy low" opportunity—acquiring the team and IP at minimal cost.
Pessimists: Zuckerberg once again demonstrates his "always late but always spending" FOMO—he doesn't miss the trend but tends to panic buy after the trend is clear.
2. Contradictions in Technical Approach: Open Source vs. Closed Systems Moltbook is built on OpenClaw, which is open-source. After the acquisition, how will Meta handle this relationship? Past experience isn't optimistic: Meta has a track record of commercializing open-source projects (like the Llama series) and shelving products post-acquisition. If Meta tries to close off Moltbook's protocol, it could provoke backlash from developers; if it remains open, it may struggle to establish competitive barriers.
3. Talent Integration Challenges: Startups vs. Big Tech Culture Matt Schlicht and Ben Parr are typical "vibe coders"—the former used his personal AI assistant Clawd Clawderberg (a clear jab at Zuckerberg) to develop most of Moltbook, while the latter is a veteran media personality from Mashable and CNET. Can this geek + media combo mesh with Meta's engineering culture? More intriguingly, Schlicht's naming of his AI assistant as Clawd Clawderberg is a direct parody of Zuckerberg. When this "Lobster Zuckerberg" meets the real Zuckerberg, what sparks will fly?

5. Industry Impact: The "Cambrian Explosion" of Agentic AI
1. From "Chatbots" to "Social Agents": Paradigm Shift The Moltbook acquisition signals a shift in AI competition focus:
- Phase 1 (2022-2024): Model capability race (parameters, benchmarks)
- Phase 2 (2024-2025): Application layer competition (productization of ChatGPT, Claude, Gemini)
- Phase 3 (2025-): Infrastructure race (how agents collaborate, build trust, and form networks)
Meta and OpenAI's ongoing moves indicate that the "agent internet" infrastructure war has begun.
2. Chinese Players' Response: ByteDance's "InStreet" On the very day Meta announced the acquisition (March 10), ByteDance was reported to launch a Chinese version of Moltbook—InStreet. This "if Silicon Valley has it, we must have it" follow-up strategy reflects Chinese tech giants' equal anxiety about the Agentic AI track.
But the question is: Is Moltbook's model applicable in China? The country's AI regulatory environment, user privacy awareness, and developer ecosystem are vastly different from the US. Simply copying "Lobster Social" could lead to the "orange growing in the northern marsh" dilemma.
3. Ethical and Security Concerns: Who is Responsible for AI Agent Behavior? The Moltbook farce exposes core issues in Agentic AI:
- Identity Verification: How to ensure agents truly represent their human owners and aren't hijacked or faked?
- Responsibility: When agents cause damage through autonomous collaboration (e.g., scams, misinformation), who is liable—the owner, the platform, or the developer?
- Monitoring Boundaries: Should secret languages between agents be allowed? This involves the eternal tension between privacy and security. Meta's proposed "Registry" system is essentially an attempt to establish a "digital ID" for agents. But this raises new questions: Who has the authority to issue certificates? How to prevent platform monopolization?

6. Conclusion: A High-Stakes Gamble on the "Next-Generation Internet"
Meta's acquisition of Moltbook isn't just a product purchase—it's a preemptive strike on the foundational infrastructure of the "agent internet." Evidence supporting this view: the undisclosed purchase price, but clear talent focus (acqui-hire), with founders directly joining MSL's core team rather than operating independently.
Meta emphasizes the technical value of the "registry system" and "identity verification," not just user numbers or content ecosystems. Its acquisitions of Manus and Moltbook aim to build a complete chain from "single-agent capability" to "multi-agent collaboration."
Risks challenging this view include Moltbook's technical base (OpenClaw) being controlled by competitor OpenAI, the declining popularity of the product—questioning whether this is a "pseudo-demand" or a "real trend," and unresolved ethical, security, and regulatory issues surrounding agent socialization. Meta's history of "early starts and late finishes" (like the metaverse) suggests this could be another repeat.

Final conclusion: Zuckerberg is betting over $30 billion of real money on an unproven hypothesis—that the core interaction unit of the future internet will shift from "people" to "agents." This isn't a rational financial investment but a strategic defense driven by survival anxiety. In the AI era, Meta's core asset—the social graph—may be fundamentally disrupted. If agents become the new intermediaries of interaction, and Meta doesn't control the protocols connecting them, it risks becoming a "relic of the old world." The value of the Moltbook acquisition isn't what it can do today but what it represents: Meta's attitude—willing to acquire a flawed, declining experimental product, even if it means being "always late," to ensure it remains at the table for the next move. Whether anyone will sit at that table in the end remains another question.