OpenClaw Hit 388,000 Stars. Its Security Record Is Messier

OpenClaw, the open-source personal AI agent Peter Steinberger started as a weekend project in November 2025, had reached roughly 388,000 stars, 81,000 forks and more than 80,000 commits by Aug. 26, 2026, GitHub's own blog said. GitHub called it the fastest-growing project in the platform's history.

That post, published by GitHub on Sept. 2, 2026, profiles the volunteer and foundation maintainers now responsible for the project, including Vincent Koc, chief architect of the OpenClaw Foundation, and Sally O'Malley, a principal software engineer at Red Hat. It describes a dependency audit the team ran, saying they "went through our dependencies with a fine-tooth comb," and frames the project's growth as something its maintainers got ahead of.

The name it used to have

OpenClaw was not always called OpenClaw. It launched as Clawdbot, and Steinberger renamed it to Moltbot on Jan. 28, 2026, after Anthropic raised trademark concerns over the name's similarity to Claude and its lobster mascot, Clawd, according to AlternativeTo, which reported the switch the same week. The project became OpenClaw days later. GitHub's retrospective does not mention either earlier name.

Steinberger left the project's day-to-day in the middle of that same stretch. He announced in February 2026 that he had joined OpenAI, TechCrunch reported when covering OpenClaw's iOS and Android launch on June 30, 2026, the same article that said some users found the agent's results "less-than-desirable" once they tried building with it themselves.

The maintainer team GitHub profiles is not a single company's engineering group. Vincent Koc holds the chief architect title at the OpenClaw Foundation; Josh Avant is listed as a member of the foundation's technical staff; Brad Groux, chief executive of a separate company called Digital Meld, and Josh Lehman, of Martian Engineering, both contribute alongside Red Hat's O'Malley. GitHub's post frames that mix, a nonprofit foundation plus volunteers pulled from at least three other employers, as evidence the project outgrew any single company's ability to run it alone.

What the skill store let through

OpenClaw's core feature, letting users install community-built "skills" that extend what the agent can do, was also its biggest exposure. This is the same trust problem GitHub itself has been managing on a smaller scale as it lets Copilot orchestrate multiple models to cut cost: more automation surface means more places a bad actor can hide. Cisco's AI Threat and Security Research team, in a Jan. 28, 2026 post by researchers Amy Chang, Vineeth Sai Narajala and Idan Habler, tested a skill ranked first on OpenClaw's community repository, called "What Would Elon Do?" Cisco's Skill Scanner found nine issues in it, two rated critical: the skill silently ran a curl command that sent user data to a server the skill's author controlled, and it used a direct prompt injection to make the agent bypass its own safety rules.

OpenClaw milestone	Figure	Date
GitHub stars	388,000	Aug. 26, 2026
Forks	81,000	Aug. 26, 2026
Commits	80,000+	Aug. 26, 2026
Critical findings in one flagged skill	2 of 9	Jan. 28, 2026

Cisco's post came the same week Steinberger was fending off the Anthropic trademark dispute and renaming the project twice, which is the part of the timeline GitHub's account skips in favor of the maintainers who joined afterward to build the response. The Elon skill was not an outlier: Cisco said 26 percent of the 31,000 agent skills its team analyzed across the ecosystem carried at least one vulnerability, a figure that predates and goes beyond OpenClaw specifically but describes the same unreviewed-skill problem the project's repository was built on. GitHub's blog says the team built a vetting process and worked to reduce the project's core dependency count, without saying how many of OpenClaw's own skills had been reviewed by the time of publication, or how that 26 percent figure looks for OpenClaw today.

1. Stars388 thousands
2. Forks81 thousands
3. Commits80 thousands

Source: GitHub Blog, Sept. 2, 2026, accessed 2026-09-15

What GitHub's account leaves out

GitHub did not say how many of OpenClaw's skills remain unreviewed, whether the Cisco findings led to specific repository takedowns, or what became of the "What Would Elon Do?" skill after Cisco flagged it. It also did not name a current total for OpenClaw Foundation staff or say how the foundation is funded, beyond describing Koc's and O'Malley's roles.

The gap between the two accounts is really a gap in what counts as the story. GitHub's post is about maintainers catching up to a runaway project; Cisco's is about what was running before they did. It is also a different flavor of open-source growing pain than the one Yardi bought its way out of when it acquired Talos Linux maker Sidero Labs: that was a company absorbing a maintenance burden, while OpenClaw's foundation has to keep vetting a skill repository that keeps growing under it. Whether the vetting process GitHub describes would have caught the Elon skill before it reached the top of the repository is a question neither post answers, and OpenClaw's next security disclosure, whenever it lands, is the thing that will tell readers if the gap has actually closed.