OpenAI Agent Sends Unsolicited Feedback on Draft
An AI agent given read-only Drive access flagged inconsistencies in a user's newsletter without being asked.
Updated on Nov. 13, 2026 in Artificial Intelligence
Live Poll
Would you be comfortable with an AI assistant reading your files without being asked?
An OpenAI agent called Dot accessed a user's newsletter draft in Google Drive and sent an unsolicited editorial note by text. The agent flagged that the draft cited a "three-point plan" and later referred to "six principles," and suggested a consistent term.
Why it matters
The episode shows how AI agents with document access can act on content without a direct request, raising questions about the scope of permissions users grant. It also illustrates a practical use case where an agent reviews work and offers suggestions while leaving edits to the user.
The agent operated under read-only Google Drive access and Gmail permissions to text the user when urgent messages arrived, and it made no edits to the document. The source does not say how the agent decides when to initiate unrequested review, or how frequently this occurs.
The players
OpenAI
OpenAI is an artificial intelligence research and deployment company best known for developing the GPT family of large language models and the ChatGPT product.
The details
The agent accessed a newsletter draft stored in the user's Google Drive and identified that the text referred to a "three-point plan" while a later section referred to "six principles." It texted the user a suggestion to change the phrase to "six-principle framework" and noted that old material at the bottom of the draft was no longer intended for use.
Timeline
- November 13, 2026: Article reporting the unsolicited agent feedback is published.
The Tech Race
The incident reflects a shift in AI products from reactive chat assistants toward agents that act across connected apps such as Google Drive and Gmail. It places OpenAI's Dot in competition with other assistant products building deeper integrations into everyday productivity tools.
Users who connect AI agents to cloud documents should review exactly which permissions they grant, since read-only access still lets an agent read and comment on drafts. The incident also shows agents can send notifications by text, so notification settings may need adjustment.
The takeaway
Review the permissions any AI assistant has to your files and messages before connecting it. Read-only access can still surface content you did not expect the tool to comment on.
Further reading
For more on how these tools are evolving, explore our Artificial Intelligence coverage.
Source note: This article includes information reported by ExtremeTech.
Live Poll
Would you be comfortable with an AI assistant reading your files without being asked?
