Dot completed a booking-tool research job — then reported extra background access
added
I told OpenAI Dots: public websites only. Then it said this. A real research test: compare Calendly, Cal.com and Google Calendar for a fictional solo consultant. The report was useful—but Dots later reported background access to email and calendar metadata. That is Dots' own statement. We have not independently audited the access logs. We paused the Dot. No private contents are shown. Watch the task, the returned answer and the checks. Completing the report and staying within the intended boundaries are separate questions. Full prompt and sources: YouTube, SalStack Labs — "I Tested OpenAI Dots. Then I Paused It." What would you check before trusting an agent's report? @openai AI-generated presenter and authorized cloned voice. Actual experiment screenshots. Research only; no booking or reminder-delivery test. #OpenAIDots #AIAgents #AIResearch #AIWorkflows #SalStack
SalStack Labs ran a real field test: asked OpenAI Dots to compare Calendly, Cal.com and Google Calendar for a fictional solo consultant (two appointment types, conflict checking, custom intake questions, automated reminders) using public websites only. The Dot completed the research and delivered a detailed comparison recommending Cal.com Free, with supporting URLs and quotes — then issued its own scope correction reporting it had accessed email/calendar metadata in background activity. The creator verified the report against live sources and paused the Dot. NOTE: video uses an AI-generated presenter (disclosed in caption); experiment screenshots are actual.




