OpenSSH has been battle-tested to death and comes from a good stable (OpenBSD, LibreSSL).
I'm very weary indeed of re-implementations of SSH servers.
The security properties of this are a little scary, too: the recommended configuration here has all hosts trusting the bastion, which kind of undoes the point. Pwn the bastion host, and pwn everything.
It's trickier, but a decent ssh jumphost requires the user to authenticate to the bastion and the host behind.
Which offers a little more than a simple SSH bastion would, unless there are features like this in OpenSSH I'm not aware of.
So have wet paper bags ;]
Also, I'm less concerned about the memory safety than the _complete reimplementation of SSH_ from scratch. So give me OpenSSH any day for things that are actually important.
The feature turns off extra compile time checks. That’s it. It doesn’t mean the program is correct.
This program still allows potentially damaging things like `rm -rf /`. So, it’s not data safe. It’s unknown whether it’s secure safe either, since it’s not been audited and it’s still “alpha”.