Not quite. Facebook still controls the endpoints, so when you see the message so can they. This is obvious: you use their app to view the encrypted message, hence the app has access to the cleartext.
If the app is not phoning home with the cleartext, this seems okay. You need some software to retrieve/read text anyway, so this becomes an exercise about trusting trust, etc.
Not at all. Good security often involves some black-and-white thinking, which not everyone is accustomed to.
If Facebook controls the endpoint, then they have the power to access the plaintext, full stop. Using their product (hopefully) implies a choice to trust them not to abuse such access.
Which is what...they said?
Would you argue against all encryption because clearly the CPU maker has a similar access to all decrypted content?
It's still better than nothing, although in the hands of Facebook it might actually be worse than nothing.
In order to actually provide your messages to Facebook, the app needs to either call home when you view the message or write the cleartext somewhere on-device to send home later. If you view the message and then the app calls out with data we can't inspect, or writes something locally that we can't inspect, it could potentially be exfiltrating the message you viewed. If not... am I missing an attack vector, or is that message safe?
(To be precise: this would only prove forward secrecy, meaning safety for that viewing of that message. If we can't see the app's code, it could have testbench cutouts like Volkswagen or WannaCry, or more likely could only trigger for certain users or in certain cases à la Greyball.)
I'm sure realistically the US gov could creatively accomplish what they want.