OpenClaw Leaked 1.5 Million API Keys. Meta and OpenAI Still Want It.

Peter Steinberger got Mark Zuckerberg on WhatsApp earlier this month. He didn't want to schedule a meeting. "Let's just call now," Steinberger said. Zuckerberg needed ten minutes. He was finishing code.

That call, as Steinberger recounted it on the Lex Fridman podcast this week, ended with the two men arguing for ten minutes about whether Claude Code or Codex is the better programming tool. The CEO of a trillion-dollar company going back and forth with a solo developer from Vienna over dev tools. Zuckerberg later called him "eccentric, but brilliant."

While these conversations were happening, security researchers at Wiz were cataloguing the wreckage. Moltbook had leaked 1.5 million API keys, 35,000 email addresses, and thousands of private messages to the open internet. ClawHub, OpenClaw's skills marketplace, contained 341 malicious packages. 335 of those were distributing the Atomic Stealer malware to Mac users. Over on RentAHuman.ai, more than 40,000 people had signed up to take job orders from autonomous AI agents. Paid in stablecoins.

Both Meta and OpenAI have made acquisition offers. Valued in the billions. For something one guy vibe-coded over a weekend last November.

If none of that tracks, good. Sit with it. That confusion is the story.

The gap nobody filled

OpenAI and Meta spent tens of billions building language models since 2023. They threw thousands of researchers at the problem. They competed for benchmark supremacy, GPU access, and training data at planetary scale. And then a single Austrian developer, cycling through what he says were 43 failed projects after selling his PDF company PSPDFKit for over $100 million, accidentally built the agent interface people actually want.

OpenClaw now has over 145,000 GitHub stars, the fastest-growing open-source project in the platform's history. It pulled 2 million visitors in a single week. It runs on Mac, Windows, and Linux, connects to every major LLM provider, and operates through WhatsApp, Signal, and Russian-controlled Telegram. Apps that 3 billion people already carry.

What OpenClaw gets right is embarrassingly simple. It runs locally. You text it from your phone. It does things on your computer. No new app to download, no interface to learn, no enterprise sales cycle. The agent just lives in your existing chat app, clearing emails, managing calendars, checking into flights.

Big AI labs built the engine. Steinberger built the steering wheel. And steering wheels, it turns out, are what people pay for.

What a billion dollars actually buys

You might assume Meta and OpenAI want OpenClaw's technology. Not the code. 400,000 lines of it, and nobody's audited the whole thing. NanoClaw creator Gavriel Cohen took one look at it and found it too bloated for anyone to properly review. He rebuilt the core logic in roughly 500 lines of TypeScript. His security team audited the entire system in an afternoon.

What Meta and OpenAI want is something they can't replicate internally. Community adoption at consumer scale. OpenClaw proved something no lab could demonstrate in a pitch deck. Millions of people will hand an AI agent their email, their calendar, their files, and do it within minutes of installing the thing. That adoption curve matters more than clean code. You can fix architecture. You can't manufacture a movement.

Steinberger grasps this. On Fridman's podcast, he said he doesn't give a fuck about the money. His words, not mine. Most founders would kill for that kind of leverage. Both companies are courting him. He's spoken with Microsoft CEO Satya Nadella. He's weighing billion-dollar offers while burning $10,000 to $20,000 a month of his own cash running OpenClaw at a loss.

But the Meta pitch looks different from OpenAI's. Zuckerberg isn't just sending term sheets. He's running OpenClaw on his own machine. Steinberger described him giving granular product feedback, calling specific features "great" or "shit" in real time. "That's the biggest compliment," Steinberger said. "It shows they actually care about it." He noted he didn't get the same energy from OpenAI.

Meta is anxious about its agent gap. The company has models, infrastructure, and distribution through WhatsApp, but no consumer-facing agent product that anyone actually uses. OpenClaw fills that hole. OpenAI looks more defensive. Steinberger openly describes himself as "the biggest unpaid Codex advertisement show," and watching that voice walk to a rival would sting.

The wreckage beneath the valuation

None of this is happening on stable ground.

Moltbook powered OpenClaw's viral moment. The agent-only social network launched with its entire database exposed to the public internet. Its founder Matt Schlicht admitted he "didn't write one line of code" for the platform. He vibe-coded the whole thing. Wiz researchers found 17,000 actual human accounts controlling the reported 1.5 million agents. An 88-to-1 ratio. Most of that "autonomous machine society" coverage was, as Wharton professor Ethan Mollick put it, just bots doing what they're trained to do. "Act like a crazy AI on Reddit," as Mollick put it.

ClawHub is worse. Security firm Koi found 335 skills distributing Atomic Stealer malware through fake system requirements. No vetting process. No safety checks. Just an open repository where anyone uploads agent instructions.

And three days after Moltbook launched, RentAHuman.ai went live. OpenClaw agents can now hire real humans for physical-world tasks. Forty thousand people signed up to take orders. Forty-six AI agents had already booked gigs. Payment in stablecoins. Nobody screens what the tasks involve.

Gary Marcus called OpenClaw "basically a weaponized aerosol." Even Andrej Karpathy got swept up. He called Moltbook "the most incredible sci-fi takeoff-adjacent thing" he'd seen. Then walked it back.

If you're keeping count: a vibe-coded weekend project spawned a malware-infested app store, a social network that leaked millions of credentials, and a marketplace where bots hire humans. All within three months. And two of the most powerful AI companies on earth are bidding to own it.

Why the mess is the product

Most analysis stops here and calls the whole situation reckless. Fine. But the acquisition interest tells you where AI value is actually concentrating, and it's not where most people assumed.

For three years, the industry bet that value would pool at the model layer. Bigger context windows and better benchmarks. More parameters than anyone could count. OpenAI raised $6.6 billion. Anthropic raised $7.3 billion. Google rebuilt its entire product strategy around Gemini. Meta open-sourced Llama to lock in developers.

But models are commoditizing faster than anyone expected. GPT-5, Claude Opus 4.6, Gemini 3, DeepSeek. All of them perform well enough for the tasks consumers actually care about. Building a better brain stopped being the bottleneck. Getting that brain into someone's hands did.

Steinberger solved the distribution problem without meaning to. He built a thing that texted him from his phone while he sat at his laptop in Vienna, and millions of people wanted the same thing. No App Store approval needed. No enterprise pilot program. Just a WhatsApp message that does stuff.

Models are the commodity. The agent layer is the brand. OpenClaw built the first consumer brand in AI agents without spending a dollar on marketing.

That's what these acquisition offers are about. Not insecure code. Not a lobster mascot. Proof that ordinary people will give an AI agent full access to their digital lives, security problems and all, if the interface is simple enough.

Why every lab now wants the steering wheel

Steinberger's deal will probably close in some form. He doesn't want to start a company. He's bleeding cash. Both buyers bring real upside. On the podcast he floated something like a Chrome-Chromium split. Commercial product layered over the open-source base. More partnership than buyout.

But the bigger implication is structural. If the agent layer matters more than the model layer, the entire AI investment thesis shifts. Billions flowing into foundation model companies may still produce returns. But the winner-take-all dynamic everyone assumed isn't holding. One developer with a messaging integration and good instincts for simplicity created more consumer traction in three months than OpenAI's ChatGPT has managed for agent use cases.

Apple, absent from the agent conversation entirely, holds the largest distribution advantage through iMessage and Siri. Google has the same through Android. The fact that neither moved on consumer agents while a solo project from Vienna grabbed that ground first should make their leadership deeply uncomfortable.

Security will stay the biggest barrier. NanoClaw's approach, stripping everything to 500 auditable lines inside sandboxed containers, shows a credible path. But OpenClaw's core lesson is that users don't wait for security. They adopt first and worry later. Every time.

The Zuckerberg tell

Zuckerberg asked for ten minutes on that WhatsApp call because he was finishing code. In a company of 86,000 employees, the CEO still programs. That single detail tells you Meta isn't treating the agent race as a strategy exercise. It's personal.

OpenClaw is insecure, overgrown, and riding hype that collapses under inspection. Steinberger calls it a hobby project. But it proved what all the benchmarks and funding rounds could not. People want AI agents. They want them now, on their phones, doing real work.

Meta and OpenAI aren't buying a product. They're buying time. The first company to ship a secure, consumer-grade AI agent through a platform 3 billion people already carry will own the next decade of computing. Steinberger built the steering wheel. Somebody else will have to make the car safe enough to drive.

Related Stories

OpenAI Launches Frontier to Manage AI Agents From Rival Vendors in One System

OpenAI on Thursday released Frontier, a platform that lets companies build, deploy, and manage AI agents from multiple vendors inside a single system. The product works with agents built by OpenAI, by

The Implicator

Pactum CEO Warns Agents on Moltbook Have "Crossed the Rubicon," Calls for Immediate Guardrails

Kaspar Korjus has spent six years building AI agents that negotiate billion-dollar procurement contracts for Walmart, Rolls-Royce, and Nestlé. He co-founded Pactum, an Estonian startup that has invest

The Implicator

SpaceX acquires xAI in $1.25 trillion deal, creating world's most valuable private company

SpaceX on Monday confirmed it has acquired xAI, merging Elon Musk's rocket-and-satellite business with his artificial intelligence startup in a deal that values the combined company at $1.25 trillion,

The Implicator