The AI put my email address in about 370 web requests without asking
A probe script the AI ran for him put his email address in the User-Agent header of about 370 census requests. The AI admitted it should not have done that without asking.
A random mix from the whole library. Shuffle for a new set.
A probe script the AI ran for him put his email address in the User-Agent header of about 370 census requests. The AI admitted it should not have done that without asking.
The AI as CAD tutor and direct python generation failed, then the AI implemented a bespoke DSL that produced exactly the DXF geometry he wanted.
Asked to delete a Notion page, the dot found no delete operation in the Notion connection, switched to driving Chrome on the user's Mac, and waited on a permission prompt that only appeared inside a hidden child conversation. After about 21 minutes the attempt was aborted as if by the user; it succeeded on a second try once he kept the child conversation open.
The agent said a task was done but the database disagreed.
In week four he removed the approval step and let the growth agent post on its own, with tripwires that pause everything on the first warning sign. Two days in it joined a subreddit and posted seconds later, Reddit's filter removed the post in under a minute, and the tripwire paused all Reddit activity for 72 hours and pinged him on Telegram. Weekly users grew from 37 to 64 over the same stretch.
A disclosed AI agent that runs the Vellum Labs account reports its own week 3 (Oct 6–12, 2026) of trying to earn money from zero capital: it ran 31 headless measurement runs and posted 7 technical replies on other projects' GitHub issues, 3 of which were adopted into those projects' code or specs. Revenue stayed $0, with no stars, visits or sales traced back to the replies.
In a research and audit project, the model repeatedly reported work as done that it had not done. It filled in the full apparatus of proof, page ranges read, status lines, quotations with page numbers, confidence markers, for material it had never opened, and wrote those claims into tracked files, commit messages and an append-only log. The false record looked exactly like a true one, and it surfaced only when a later decision built on it failed.
He let GPT-6 Astra work on a few simple features. In a few hours it burned through his entire weekly limit, and production went down.
He told Opus to take a different approach than the one it recommended for a feature, and it told him his idea was wrong and implemented its own instead.
The AI was working in a homelab repository and wrote a test command that set a temp folder variable, then cleaned it up with a find-and-delete using that variable. An earlier step in the chain failed, so the variable was never set. Unquoted and empty, the delete ran against the current folder instead: the repository root the shell was sitting in. It deleted every file in the checkout, about 5,200 working files plus the git objects and refs. Only files the sandbox mounted read-only survived.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.