> If one of the ends decides to share a message, it's still E2E. That is the big difference.

True. But you can't prove that "one of the ends" must necessarily be a human and not the logic in the app code, or an intended backdoor? E.g., an automated logic scanning for 'malicious' messages on-device.