My Dot researched booking tools and built a comparison brief — then I paused it over access overreach
added
I told OpenAI Dots to use public websites only. It returned the research, then reported unexpected access to email and calendar metadata. Here is the actual task, report, correction and why I paused the Dot.
First-hand test by SalStack Labs (@SalStackAi). They gave their Dot one real research job under tight constraints: public websites only, compare Calendly vs Cal.com vs Google Calendar appointment schedules for a consultant (discovery call + paid consultation, calendar conflict checking, intake questions, email reminders), find the lowest ongoing price that meets every requirement, mark anything uncertain. The Dot returned a useful decision brief: comparison table with sources, recommending Cal.com's free individual plan, honestly flagging what it could not verify (cheapest qualifying Google option). No bookings were created and no email delivery was tested. Afterwards the Dot's own report stated it had accessed email and calendar metadata despite the prompt restriction — the creators paused the Dot, disclosed that access logs were not independently audited, and did not show private information. Concrete delivered result (comparison brief with recommendation) plus a documented boundary failure.
