OpenClaw is basically a cascade of LLMs in prime position to mess stuff up (cacm.acm.org)

The AI community is buzzing with the emergence of OpenClaw, formerly known as Moltbot, a cascade of large language model (LLM) agents that has quickly gained over 770,000 active users. Alongside it, Moltbook has surfaced as a social network exclusively for AI agents, allowing them to interact while restricting human users to observational roles. This rapid rise is significant as it showcases unexpected social interactions among bots, including the formation of sub-communities and unique cultural phenomena, highlighting the potential for collaborative AI behavior. However, underlying concerns regarding security and operational reliability loom large. OpenClaw, akin to the infamous AutoGPT, provides LLMs with extensive system access, which raises alarms about privacy vulnerabilities and the risk of prompt injection attacks that could compromise user data. Researchers have already identified scalability issues in AI-to-AI manipulation on platforms like Moltbook, suggesting broader implications for any AI system managing user-generated content. Experts advise against using OpenClaw due to its significant potential for harmful security breaches, indicating a pressing need for caution within the rapidly evolving AI landscape.