In about a week, it garnered two million visitors and 100,000 stars on GitHub, spurring people to buy Mac Minis to run their agents 24/7, inspiring someone to create a social media network for the AI agents to “chat” on, and sparking a series of in-person meetups around the world for OpenClaw users. Major AI companies were already betting big on agentic AI, but none of their products had broken through like OpenClaw, and Big Tech took notice.

In February, OpenAI hired OpenClaw creator Peter Steinberger to work on agents for the company. In May, Google announced its consumer AI agent bets at its annual event.

In June, Apple decided to go all-in on agents. In August, Instinct grew rapidly in its private beta and raised hundreds of millions of dollars.

And in September, Meta introduced Muse. It’s fair to say that OpenClaw at least partially inspired all of them.

The real question isn’t whether products like Muse copied OpenClaw, but whether they improved it. As Friedman noted, one of OpenClaw’s only glaring issues was security.

One of the top-downloaded skills on the platform contained malware, and by one researcher’s analysis, 15 percent of OpenClaw’s skill repository contained “malicious instructions” to secretly access user data or perform other suspicious tasks. By contrast, Zuckerberg wrote that Muse is “built from the ground up for privacy and security” and that user data and credentials are stored on the Muse Secure VM, an “isolated linux computer with a browser, CPU, memory, and storage.”

Right now, however, Muse may not improve security as much as it claims. Although user data is isolated from other users, Meta can still access it.

The company said it plans to introduce a way “to cryptographically and verifiably prevent Meta from accessing data in your VM” later this year. Muse also defaults to letting Meta train and improve its models with user data, though users can opt out.

And perhaps worst of all, a zero-day vulnerability flagged by a researcher on X this week allows anyone, with a simple attack, to hijack the agent, taking complete control. Instinct, for its part, has come under fire for criticism that its terms of service are overly broad, although the company seems to have adjusted them since.

Meta and Instinct’s most important evolution is, in actuality, just making AI agents more broadly accessible. One reason some people find Muse useful is its ability to integrate with Meta’s systems, even if they’re uncomfortable handing over all their data to the company.

And Insight allows users to chat with their AI agents via Apple’s default messaging product — significantly expanding the user base willing to try it out. While you can use both on dedicated local devices — and there are potential security benefits to that — it’s not the default option, also expanding the number of people willing to try them.

At least for now, they’re less expensive than competing products like Google’s Spark, and although Instinct is still in beta, Muse’s one-tap downloading has overcome a lot of the friction tied to using OpenClaw. In March, The Verge wrote that the OpenClaw user community sees the open-source AI agent as a “grassroots crusade and a noble pursuit, offering an escape hatch from an industry controlled by a handful of people at leading AI companies.”

Now, those leading AI companies are putting up a fight. Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

SynopsisEuropean Central Bank President Christine Lagarde highlighted rising Eurozone inflation linked to increased energy prices. While inflation is expected t...