AI researcher Gary Marcus sounds off on Moltbook and OpenClaw's viral moment

Gary Marcus, a prominent AI researcher and author, has voiced significant concerns regarding the recent surge of viral AI agents, particularly OpenClaw and Moltbook. Last week, animated lobsters took over social media platform X, igniting discussions about these emerging technologies. OpenClaw, which has undergone several name changes, operates locally and can autonomously interact with various consumer applications without human oversight. Meanwhile, Moltbook functions as a Reddit-like forum exclusively for AI agents, where human participation is ostensibly barred, although some individuals may have circumvented this restriction. Marcus, known for his cautious stance on AI advancements, stresses the importance of considering potential risks associated with these tools. In a recent Substack post, he stated unequivocally, "If you care about the security of your device or the privacy of your data, don't use OpenClaw. Period." In an email exchange, Marcus elaborated on his initial encounter with OpenClaw, recalling his immediate reaction to its similarities with AutoGPT. He warned that if it gains traction, it could lead to serious security issues. He speculated that while users might find it entertaining and futuristic, he would advise against engaging with it altogether. When comparing OpenClaw to AutoGPT, Marcus noted both serve as bridges between large language models (LLMs) and external services, but the security risks remain substantial. He likened the situation to handing over your computer and passwords to a stranger in a bar—an inherently dangerous proposition. As for those who have already implemented OpenClaw, Marcus expressed doubt about any secure usage. He recommended uninstalling the application and updating passwords for sensitive accounts. Although OpenClaw's creator, Peter Steinberger, claims to have enhanced its security, Marcus likens these efforts to a temporary fix rather than a comprehensive solution. Turning to Moltbook, Marcus questioned the rationale behind deploying agents on the platform, referencing a proposal from startup advocate Matt Schlicht, who suggested building businesses on Moltbook. Marcus cautioned that such ventures could expose users to prompt injection attacks and other risks. Meta’s CTO, Andrew Bosworth, also commented on Moltbook, expressing disinterest in its premise as the agents are trained to mimic human speech. Marcus agreed, citing past experiments by Facebook but emphasizing the novelty of the scale. As for the future of Moltbook, Marcus remains skeptical. Drawing parallels to the rapid rise and fall of AutoGPT earlier this year, he speculated that Moltbook could also fade into obscurity, akin to other fleeting internet trends. Only time will tell if these AI agents are here to stay or just another passing fad.