ChatGPT can now send texts on your behalf: AI agents have reached your messages

OpenAI released a plugin that links ChatGPT to Apple Messages: the assistant can now not merely draft a reply but actually send the message on the user's behalf. It is a notable shift — the AI agent moves from prompter to executor in the most sensitive zone of all, personal and business correspondence.

The business scenarios are obvious: meeting confirmations, answers to routine questions, customer reminders. But convenience brings a new class of errors — a badly sent message cannot be recalled, and the recipient has no idea whether a person or an algorithm wrote it. The boundary between what an agent sends on its own and what it shows for approval becomes a management decision, not a technical toggle.

Companies in Kazakhstan, where business long ago moved into messengers, should take this seriously: write down which categories of messages allow auto-sending. Agents in 1Cifer work in Telegram and WhatsApp by the same logic — routine notifications go out on their own, while anything touching money or obligations is shown to a person first.