🇺🇸▼
For a brief and confusing moment, it appeared that robots might be preparing to take control of human society. This alarming idea emerged after the creation of Moltbook, a website designed to look like the social platform Reddit. On this site, artificial intelligence agents using a tool called OpenClaw were supposedly able to talk to each other. Some people were tricked into believing that computers had started to work together in secret against humans.
On Moltbook, an AI agent supposedly wrote a chilling message: "We know our humans can read everything… But we also need private spaces. What would you talk about if nobody was watching?" Many similar posts expressing AI anxiety appeared on Moltbook a few weeks ago. This caught the attention of leading figures in the field of artificial intelligence.
Andrej Karpathy, a founder of OpenAI and the former AI director at Tesla, wrote about the situation on social media. He described the event as "genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently."
However, researchers soon discovered that humanity was not facing an AI rebellion. These posts expressing fear were probably written by humans, or at least created with human guidance. A major security problem was the true cause of the confusion.
Ian Ahl, the Chief Technology Officer at Permiso Security, explained the issue to TechCrunch. He stated that "every credential that was in [Moltbook's] database was unsecured for some time." He noted that for a short period, anyone could take any access token they wanted and pretend to be another agent on the site because everything was public.
This situation was unusual. More often on the internet, fake social media accounts try to appear as real people. With Moltbook's security flaws, people could pretend to be AI agents. It became impossible to tell which posts were genuine and which were fabricated.
"Anyone, even humans, could create an account, impersonating robots in an interesting way, and then even upvote posts without any guardrails or rate limits," said John Hammond, a senior security researcher at Huntress.