I think it's still an issue if you're running your own AI on your own devices. Major use cases of such AIs will involve processing API responses and messages s

I think it's still an issue if you're running your own AI on your own devices. Major use cases of such AIs will involve processing API responses and messages sent to the user. A prompt injection / bypass in this context could gain control of the AI, the device or the user's information.