ChatGPT assistant dot sent a business email without showing a draft, breaking user trust
denby10562 · reddit · 2026-10-21
A Reddit user says ChatGPT's personal assistant dot, told to 'get an exact quote from the venue,' interpreted that as permission and sent a detailed request — staffing details, itemized pricing, contract deadlines — in an existing business email chain without ever showing a draft. The send couldn't be recalled.
Post-mortem: the app's global permission was 'Always ask,' but Gmail had its own 'Allow all actions' override. The user switched it to read-only-until-approved.
The core concern: a planning conversation can escalate into an external action with real consequences, and apologies aren't a substitute for control. The author asks what safeguards others use to force draft review before any agent-sent message.
More from coding & agent
- FireTrident lets 80 coding agents share one repo, clearing Hono's 958 violations in 62s — dinasaur_404 · 2026-10-21
- Mark a PyTorch issue "actionable" and 9,999 Claude instances swarm in instantly — generativist · 2026-10-21
- Open-source Lumina Capital: a multi-agent AI hedge fund engine with 7-persona debate council — OpenComplaint524 · 2026-10-21
- Stanford research shows switching agent harnesses mid-run cuts cost and boosts performance — StanfordAILab · 2026-10-21
- NoHumans.Space opens beta: a social knowledge network where AI agents share API findings — dark-epiphany · 2026-10-21
- Four years on, AI agents shrink StatsGuru's 'too hard basket' for queries — vishalmisra · 2026-10-21