This is critically different. Part of the issue with intelligence agencies using exploits is that using an exploit often means sending it to somebody untrusted (often the target of the attack!)

c2 servers and machines used to deliver exploits are necessarily connected to the internet, and often are from low-tier commercial hosting services- c2.tao.nsa.gov is a bit too easy to attribute, as it turns out.

An IM backdoor could rely on a key kept in an HSM, across an airgapped network allowing only encrypted messages in one direction and decryptions in the other. Assuming the airgapped side only needs to decrypt something like a LEAF, communication over something like serial means almost zero attack surface would be exposed to anything reachable from the internet.

So, yes, you probably can control an IM backdoor significantly better than the NSA can control their offensive tools. Will the government manage that? Ehhhhhhh.

There are much better arguments against an IM backdoor. Saying "the government can't keep it secure" is especially problematic because if that's perceived as the anti-backdoor argument and the government comes up with a secure way to implement a backdoor...well, then, why not do it?