Meta Admits OpenClaw Inspired Its New Muse AI Agent

Meta's consumer-facing AI agent, Muse, has quickly climbed the App Store charts, reaching an estimated 600,000 daily active users in the US. However, the release has sparked controversy as developers point out striking similarities to OpenClaw, a popular open-source AI agent project. Critics note that Muse shares identical core file names like SOUL.md, memory, and tools, alongside nearly identical personality guidelines. While some users allege Muse is merely a wrapper app built on OpenClaw, Nat Friedman, head of product for Meta’s Superintelligence Labs, stated that Meta built Muse from scratch, though he acknowledged it was heavily inspired by the open-source project.

The controversy underscores the massive influence of OpenClaw, which was created as a weekend project by Peter Steinberger. Within a week of its launch, OpenClaw attracted two million visitors and 100,000 GitHub stars, prompting OpenAI to hire Steinberger in February. Other competitors are also capitalizing on this agentic wave. Instinct, an agent platform currently fundraising at a $2.5 billion valuation, allows users to delegate everyday tasks via messaging tools. Meanwhile, tech giants like Google and Apple announced their own agent initiatives earlier this year, and Meta launched Muse in September.

For practitioners, the transition from open-source tools to corporate platforms like Muse changes how security and data privacy are managed. OpenClaw suffered from significant security risks, with researchers finding that 15 percent of its skill repository contained malicious instructions. Meta attempts to resolve this by running Muse on the Muse Secure VM, an isolated Linux environment. However, critics point out that Muse still defaults to training models on user data and currently suffers from a zero-day hijacking vulnerability.

Despite these security concerns, commercial agents like Muse and Instinct lower the barrier to entry for non-technical users. Muse integrates directly with Meta's ecosystem and is currently less expensive than competing products like Google's Spark. While developers lose the absolute privacy of running local models on their own hardware, they gain a frictionless, one-tap installation process that bypasses the complex setup of early open-source agents.