All AI news

OpenAI explains agent monitoring: a pause is not proof of user misconduct

Automatic API stopping has conditions and does not undo completed actions.

OpenAI described its agent-monitoring system on October 29. For covered models, tool-using Responses API requests are monitored; automatic stopping additionally requires supported conversation tracking. Chat Completions is outside this system; other safety checks still apply.

An alert calls for review, not a finding of user misconduct. Asynchronous monitoring can miss issues or raise false alarms, and cannot undo earlier actions. The publication date does not establish same-day activation for every account. Not independently tested.

https://developers.openai.com/api/docs/guides/safety-checks/misalignment-monitoring
https://help.openai.com/en/articles/20001509-why-was-my-chat-paused-or-ended-as-a-precaution
https://help.openai.com/en/articles/5347006-openai-api-supported-countries-and-territories
https://help.openai.com/en/articles/7947663-chatgpt-supported-countries

Usage conditions

Hong Kong
Hong Kong is absent from the official ChatGPT web/mobile and API support lists. The monitoring announcement is not regional access; local plans, prices and Traditional Chinese notices are unconfirmed.
Macau
Macau is absent from the official ChatGPT web/mobile and API support lists. The monitoring announcement is not regional access; local plans, prices and Traditional Chinese notices are unconfirmed.
Taiwan
Taiwan is listed for ChatGPT web/mobile and the API. Monitoring still depends on model and surface; account-plan eligibility, separate monitoring charges and Traditional Chinese notices are unconfirmed.

Official source: How we detect frontier model misalignment

Source published:

Source checked:

Published here: