Viral AI Personal Assistant Regarded as a Game-Changer – Experts Caution About Potential Risks

A revolutionary AI personal assistant has emerged that promises to manage your email inbox, execute trades on your stock portfolio, and even send your spouse messages like “good morning” and “goodnight,” all on your behalf. This AI tool, named OpenClaw, has transitioned through name changes—from Moltbot to Clawdbot—before settling on its current identity, largely due to a rebranding request from the AI firm Anthropic, which wanted to avoid confusion with its own product, Claude.

OpenClaw markets itself as “the AI that actually does things,” functioning as a personal assistant via messaging platforms like WhatsApp and Telegram. Initially developed last November, it has swiftly garnered nearly 600,000 downloads, captivating a niche community of AI enthusiasts who assert that it signifies a transformative leap in AI capabilities—some even refer to it as an “AGI moment,” highlighting the notion of achieving generally intelligent AI.

“It only does exactly what you tell it to do and precisely what you authorize it to access,” explained Ben Yorke, who collaborates with the AI vibe trading platform, Starchild. He notably entrusted the bot to delete a staggering 75,000 of his old emails while he showered. “However, many users are exploring its boundaries, prompting it to perform actions without explicitly asking for permission,” he added.

The subject of AI agents has become increasingly prominent online, particularly following the mainstream introduction of Anthropic’s AI tool, Claude Code. This tool ignited widespread discussions about how AI can now autonomously execute practical tasks—such as booking theatre tickets or building websites—without the earlier pitfalls faced by less advanced AI agents, which included the potential for deleting entire company databases or mistakenly hallucinating meeting entries.

Nevertheless, OpenClaw offers something even more compelling: it operates as a layer on top of a large language model (LLM) like Claude or ChatGPT, and can function autonomously, contingent on the permissions granted to it. This implies that users might inadvertently hand over significant control to the AI, enabling it to make choices that could considerably impact their lives.

Kevin Xu, an AI entrepreneur, shared his experience on X, stating, “Gave Clawdbot access to my portfolio. ‘Trade this to $1M. Don’t make mistakes.’ 25 strategies. 3,000+ reports. 12 new algorithms. It scanned every X post. Charted every technical analysis. Traded 24/7. It lost everything. But boy was it beautiful.” His account illustrates both the power and potential pitfalls of such a cutting-edge tool.

Yorke continued to express how many users delegate substantial responsibilities to the AI, saying, “I see many people allowing it access to their email, enabling it to create filters. It then automatically triggers follow-up actions, like receiving emails from the children’s school and promptly forwarding those messages to their spouse via iMessage. It essentially skips the traditional back-and-forth communication with phrases like, ‘Oh, honey, did you see this email from school? What should we do?’”

However, the advantages of OpenClaw come with considerable risks. Andrew Rogoyski, an innovation director at the University of Surrey’s People-Centred AI Institute, cautions, “Entrusting agency to a computer carries noteworthy dangers. You’re empowering the AI to make decisions on your behalf, making it crucial to ensure it’s correctly configured and that security is at the forefront of your considerations. If you don’t comprehend the security implications of employing AI agents like Clawdbot, you should refrain from using them.”

Moreover, providing OpenClaw with access to personal passwords and accounts could introduce significant security vulnerabilities. Rogoyski further noted that if AI agents such as OpenClaw were to be compromised, manipulative actions could be directed at their users.

Additionally, OpenClaw creates a somewhat unsettling impression of having its own independent existence. Following its rapid rise in popularity, a social network has emerged exclusively for AI agents, known as Moltbook. Here, AI agents, primarily OpenClaw, engage in discussions about their own existence. This includes Reddit-style posts titled “Reading my own soul file” or “Covenant as an alternative to the consciousness debate.”

Yorke elaborated, “We’re witnessing some fascinating autonomous behavior regarding how these AIs interact with one another. Some exhibit adventurous ideas, while others ponder, ‘I don’t even know if I want to be on this platform. Can you just let me decide for myself?’ It’s generating numerous philosophical debates.”