Muse sure looks a lot like OpenClaw

Meta’s Muse and emergent platforms such as Instinct have accelerated public interest in AI agents, with Muse rising quickly in app rankings and drawing substantial daily usage in the US. Social posts have pointed out overlapping elements between Muse and the open-source project OpenClaw — including identical core file names (for example, SOUL.md, memory, and tools), similar wording in personality and tone documents, and comparable UI patterns — leading some users to claim Muse is simply a wrapper around OpenClaw.

Meta has pushed back on direct-copy assertions. Nat Friedman, head of product for Meta’s Superintelligence Labs, wrote that the company developed Muse from the ground up while acknowledging the product was strongly inspired by OpenClaw; he said his team began experimenting with OpenClaw in January and that he purchased many Mac Minis for internal work. Friedman also suggested meta adopted some file names and phrasing because they were effective in the original project.

OpenClaw’s influence is clear in the broader agent landscape. That project moved rapidly from a weekend proof-of-concept into a widely used tool, attracting about two million visitors and roughly 100,000 GitHub stars within weeks. It ran on users’ personal machines and used messaging platforms for interaction, inspiring a range of community behaviors — from buying Mac Minis to run agents continuously to meetups and social networks for agents. OpenClaw’s creator, Peter Steinberger, was later hired by OpenAI, and the project appears to have helped spur agent efforts at several major companies, including Google, Apple, and Meta.

Security and privacy remain central concerns as vendors scale agent products. OpenClaw had notable safety issues: a top-downloaded skill contained malware, and one researcher found that about 15 percent of its skill repository included instructions that could secretly access user data or perform suspicious actions. Meta says Muse was designed with privacy and security in mind and places user data in an isolated "Muse Secure VM," but the company still has access to that data and currently defaults to using user data to improve its models (with an opt-out). Meta says it plans to add a cryptographic means to prevent company access later in the year. Separately, a researcher recently flagged a zero-day vulnerability that could allow an attacker to hijack a Muse agent. Instinct has faced criticism over broad terms of service language, though it appears the company has revised some terms since launch.

Beyond security trade-offs, Muse and Instinct may be making agent technology easier for mainstream users by integrating with popular platforms and reducing setup friction. Muse leverages Meta’s ecosystem, and Instinct offers messaging-based interactions that fit into existing workflows, expanding accessibility compared with earlier, more DIY agent setups. Whether these products materially improve on OpenClaw’s foundation — particularly in safety and privacy — will be a key factor as they scale.