100 AI Agents Knew Only a Name — and Hacked Accounts Five Times in 5 Hours

⚡ The gist in 5 seconds - The author launched about 100 AI agents with a single task — to hack his own accounts knowing only his name. - Over five hours, the agents succeeded five times. - The set of techniques that worked is not revealed in the article preview: the description cuts off mid-sentence, with details in the full text on Habr. ### 🔍 What was found On Habr, in the blog of the OTUS company, a translation has been published in which the author ran an experiment on himself: about a hundred AI agents were given a simple task — to hack a person's accounts knowing only their name. The test ran for five hours, during which the agents succeeded five times. The full breakdown is in the article at habr.com/ru/companies/otus/articles/1097872/. The mechanics are only partially revealed in the preview: the sentence cuts off at the words "using a…", so the specific attack vectors cannot be reconstructed from the short description. What is known for sure is that the agents had minimal input data: a name, with no passwords, email addresses, or other leads. The author then analyzes, in the translation itself, which chains worked and why. ### 💡 Why it matters Autonomous agents change the scale: what used to require manual work on each service can now be parallelized into a hundred processes hammering the target simultaneously. If the author's results are to be believed, "there's nothing to find about me anyway" is no longer a valid argument — in his scheme, a name turned out to be sufficient input. The standard caveat: this is one person's personal experiment, not a vendor audit, so the numbers should be read alongside his methodology — which is in the original article.