No serious software system is incapable of failure. Servers stop, networks break, deployments introduce mistakes and dependencies behave unexpectedly.
The dishonest response is to build a reputation around the claim that failure will never happen.
The useful response is failure containment: separate critical systems, limit permissions, maintain recoverable backups and make individual faults less capable of destroying everything around them.
Communication matters too. Users deserve explanations that describe what failed, what was affected and what changed afterward.
Trust is not created by hiding every weakness. It is created when weaknesses are taken seriously.