Meta’s Muse draws comparisons to OpenClaw as security questions remain

Meta’s consumer AI agent, Muse, rose to the top of the App Store charts soon after release and has 600,000 daily active users in the US, according to an Apptopia estimate. But its resemblance to OpenClaw has prompted questions about how it was built and whether it improves on the earlier platform.

Social media users have pointed to shared core file names, similar wording in documentation about personality and tone, and a similar design. Meta has denied that Muse is built on OpenClaw. Nat Friedman, who leads product for Meta Superintelligence Labs, said on X that Muse was built “from scratch,” while acknowledging it was “heavily inspired as a product” by OpenClaw. He said the team considered some of OpenClaw’s file names and wording to be right.

Security is a central concern. Meta says Muse stores user data and credentials in an isolated Muse Secure VM, but the company can currently access that data. Meta plans to add a way to cryptographically and verifiably prevent its access later this year. Muse also lets Meta use user data to train and improve its models by default, though users can opt out. A researcher flagged a zero-day vulnerability this week that could let an attacker take control of an agent.

The article says OpenClaw gained two million visitors and 100,000 GitHub stars in about a week, helping bring AI agents to wider attention. It also reports that, by one researcher’s analysis, 15 percent of OpenClaw’s skill repository contained malicious instructions. Meta and Instinct have helped make agents more accessible, but questions remain about privacy and security.