Hi,
General issue with the Gmail integration/“connector”.
I use it routinely as a way to ask ChatGPT to send me and my wife a notification message. For example, after a useful chat, I might ask ChatGPT to create a document using the drive connector and then send us an email with a link to the new doc.
I allowed the tool use request (which I think should have an always allow option!) But even after doing that, the tool use chokes on a safety model that seems to get invoked after the tool use is authorized.
I can see the PII detection flagging for a second, with things like contains name, email address.
But… That’s kind of the point!
The way it’s set up currently makes it kind of a crapshoot as to whether whether any request to use the tool will actually work or result in a rejection, which then seems to trigger some kind of long-lasting standoff, such that subsequent requests, even when the potential triggers are removed prevents the tool used from going through again.