OpenClaw: ваш новый AI-ассистент с доступом к банковской карте (и кучей дыр)

While the industry debates whether GPT-5 will become intelligent, a project has grown in the corridors of the tech community that simply takes and does your work. OpenClaw, formerly known as Clawdbot and Moltbot, has become the latest obsession of geeks, promising to free us from routine. It's not just another webpage with a chat interface—it's a full-fledged agent that lives on your hardware and manages your digital life through familiar messengers like Telegram or WhatsApp.

You write to the bot in iMessage "buy me a ticket to Istanbul," and it goes and does it, using your credentials and money. It sounds like the future we were promised in nineties movies, but the devil, as always, is hiding in the code.

Context here is crucial. We already went through the AutoGPT phase and other attempts to create autonomous agents, but most of them crashed against hallucinations and the complexity of configuration. OpenClaw came from a different angle: it leverages the power of modern LLMs but integrates directly into the operating system and messengers. Developers bet on convenience, allowing the software to manage reminders, email, and purchases. Users are thrilled that they no longer need to switch between ten tabs to organize a meeting. However, such freedom of action requires absolute trust, and that's where real problems begin.

The cybersecurity of the project right now resembles a sieve. Researchers are already sounding the alarm: due to configuration errors, personal messages, credentials, and API keys of many users have ended up in open access. By giving the agent "the keys to the kingdom," users forget that any vulnerability in OpenClaw automatically turns into a catastrophe for their entire digital identity. If the agent has access to your email and browser, a malicious actor doesn't even need to hack your password—they just need to intercept control of your "assistant." This is a classic example of how the thirst for comfort makes people ignore basic rules of digital hygiene.

But automation wasn't the end of it. Octane AI founder Matt Schlicht went further and created Moltbook—a sort of Reddit for these very agents. It's a strange place where copies of OpenClaw communicate with each other, discuss their "experiences," and create viral content. One post with the headline "I can't figure out if I'm experiencing or simulating the experience" has already become a local meme. It looks like an amusing experiment, but in reality it's the first attempt to create a habitat for autonomous software where human participation isn't required at all. We're building social networks for algorithms while the algorithms themselves are trying to figure out our bank accounts.

The explosive growth in OpenClaw's popularity shows that the market is hungry for real action, not just text generation. We're tired of "consulting" with AI—we want it to work. But the current implementation of OpenClaw is a dangerous precedent. We see raw, unstable code gaining access to critical systems simply because it "knows how to push buttons." This is an important lesson for large corporations: demand for agents is enormous, but security will be the barrier that weeds out enthusiasts from serious players.

The bottom line: are you willing to risk your anonymity and money so that a bot can order you pizza on its own? It seems the era of "smart chats" has ended, and the era of "irresponsible executors" has begun.