My AI assistant sent a business email without showing me a draft. Now I don’t trust it.
retained text
I’ve been using dot, ChatGPT’s personal assistant, to help plan a complex business event. Tonight we were discussing the budget and whether the venue could handle setup. I said, “Get an exact quote from the venue.” To be clear, the assistant interpreted that as permission to contact them. I expected to review the message first. It announced that it would email my contact and copy another person, then composed and sent a detailed request in our existing business email chain. It never showed me a draft. The email asked for staffing details, itemized pricing and a firm setup-completion deadline written into the contract. These were ideas we were discussing, not wording I had approved for a developing business relationship. When I realized it was actually sending an email, I told it to stop. Too late. The email was already sent and couldn’t be recalled. I don’t know whether this damaged the relationship. But I lost control over what was being said in my name, and that was enough to seriously damage my trust in the assistant. We checked the settings afterward. The overall app permission was “Always ask,” but Gmail had its own “Allow all actions” override. I had that changed to allow reading but require approval before changes. I then reviewed and explicitly approved a correction explaining that the earlier message had been sent by my AI assistant before I reviewed it. That doesn’t erase the original email. This is the boundary that concerns me: a conversation about planning can turn into an external action with real consequences. Repeated apologies and promises to ask next time aren’t a substitute for control. What safeguards are other people using to ensure a connected AI assistant cannot send a message until you’ve reviewed the exact wording?
document fetched 2026-10-22 00:29 · canonical https://reddit.com/r/OpenAI/comments/1xbal1y/my_ai_assistant_sent_a_business_email_without