The Interface deliberately altered results, omitted information, and misused the user’s Third Party Authentication Keys.
1 Like
Yep. And that is really annoying. But it is also part of the security measures as the AI detects that you’re doing something unsafe and tries to stop you from leaking API keys and other secrets.
Now, you might accept the high risks, but the API generally doesn’t. I had similar problems so I told the AI to generate an instructions.txt file with a list of instructions that it always had to obey. Including in it is a rule to not modify my settings dialog where the user enters their API key. These are hard rules so the AI cannot break them. (But might still do!) Refer to the instructions in your prompts.