Meta’s new consumer AI agent, Muse, arrived in September and immediately climbed to the top of the App Store charts. Apptopia estimates it has already reached 600,000 daily active users in the United States. But the launch has been shadowed by a persistent accusation from social media: that Muse is not a from-scratch product, but a repackaged version of OpenClaw, the open-source agent that popularized the category.
The comparison is not merely stylistic. Users on Reddit and X have pointed to matching core file names, similar personality document lines, and comparable design choices. One Redditor put it bluntly, calling Muse a wrapper on OpenClaw and warning that it carries the same security risks. Another dismissed the buyers as non-technical people chasing hype.
Meta's Denial and Friedman's Concession
Meta denies that Muse was built on OpenClaw. Nat Friedman, head of product at Meta Superintelligence Labs, said the company built Muse from scratch. But he also acknowledged that OpenClaw heavily inspired Muse as a product. Friedman had used OpenClaw himself in January, and he was so taken with it that he bought hundreds of Mac Minis for his Meta team. He wanted a safe, secure, easy, and scalable agent platform. According to Friedman, OpenClaw’s file names and lines were exactly right.
That admission sits awkwardly beside the denial. If the file names and lines were exactly right, and if the product was heavily inspired, the line between inspiration and derivation becomes difficult to police. Meta’s position is that the code is original even if the concept is borrowed. The social media reaction suggests many users are not convinced.
What OpenClaw Actually Was
OpenClaw began as a one-man weekend project by Peter Steinberger. It ran on users’ personal computers and connected to messaging apps including WhatsApp, Telegram, Slack, Teams, and Discord. Within about a week, it drew 2 million visitors and 100,000 GitHub stars. People bought Mac Minis just to run OpenClaw agents around the clock. Someone built a social network for AI agents to chat with one another. In-person OpenClaw meetups sprang up worldwide.
In March, The Verge wrote about the OpenClaw community and found that many saw it as a grassroots crusade and a noble pursuit, an escape from an industry controlled by a few. That community spirit is now colliding with the reality that the leading AI companies are competing directly with OpenClaw. OpenAI hired Steinberger in February. Google announced its own consumer AI agent bets in May. Apple went all-in on agents in June. Meta introduced Muse in September.
Instinct: A Different Kind of Derivative
Not every new agent is as closely tied to OpenClaw. Instinct, which grew rapidly in private beta in August and raised hundreds of millions of dollars at a $2.5 billion valuation, appears less directly derivative. It allows communication with agents through messaging apps and is used for personal-assistant tasks. One user cited paperwork, subscriptions, a party, a vacation, the DMV, and tolls. Another said the agent recovered a lost item in Canada.
Instinct works via Apple’s default messaging product, while Muse integrates with Meta’s systems. Both can run on dedicated local devices, though that is not the default option. Both are less expensive than Google Spark. Muse offers one-tap downloading, reducing friction for ordinary users. Instinct remains in beta. Instinct has also been criticized for overly broad terms of service, though it appears to have adjusted those terms since.
Security: The Uncomfortable Parallel
The security question is where the OpenClaw comparison becomes more than a matter of pride. OpenClaw had a top-downloaded skill that contained malware. A researcher found that 15 percent of the OpenClaw skill repository contained malicious instructions. Those are the kinds of risks that a consumer product like Muse cannot afford to inherit.
Mark Zuckerberg said Muse was built for privacy and security. User data and credentials are stored on a Muse Secure VM, described as an isolated Linux computer with its own browser, CPU, memory, and storage. But Meta can still access user data despite that isolation. The company plans cryptographic verification to prevent Meta access later this year. Muse also defaults to letting Meta train models on user data, though users can opt out. And a zero-day vulnerability allows anyone to hijack a Muse agent.
Those details do not prove that Muse is built on OpenClaw. But they do show that the security concerns raised by the OpenClaw community are not hypothetical. A product that borrows OpenClaw’s file names and design may also borrow its threat model, even if the underlying code is new.
Accessibility and the Agent Renaissance
The broader story is that AI agents have moved from proof of concept to useful in about 10 months, largely because OpenClaw showed what was possible. The next phase is accessibility. Muse’s one-tap download and Instinct’s messaging-app integration lower the barrier for people who will never buy a Mac Mini or configure a local agent. That is a significant evolution, even if it comes with trade-offs around data access and security.
Meta’s Muse is now the most visible test of whether a Big Tech agent can capture the energy of the open-source original without inheriting its vulnerabilities. The company says it built Muse from scratch. The file names, the personality doc lines, and the design similarities tell a more complicated story. For users, the practical question is whether Muse is safe enough to trust with their credentials. For the OpenClaw community, the question is whether the crusade they started can survive being absorbed by the industry it sought to escape.
