Here's another way of looking at it: Many mutations are going to happen in the wild, but only a few will make the virus more viable in some host or hosts. Most of the mutations will be dead ends, or jump to another species and then mutate away. Only rarely could one emerge and jump all the way to humans, spread across the globe and cause a pandemic. Ebola couldn't, and SARS1 as well due to speed of mutation. (The only original SARS1 around any more is in laboratories, except when it has escaped.)

In GOF you speed that process up by orders of magnitude. You arrive at the viable mutation sooner, possibly infinitely sooner if random mutation would never have arrived. If you engineered part of it, there's a good chance it would never have emerged naturally. It might have come into the world in a few decades or centuries, or it might never have. But you have it now, and have to guard it perfectly as long as it remains in your lab.

Bottom line, it probably won't escape, but it probably never would have emerged naturally either. IMO, to make GoF a beneficial trade-off, it should be focused on viral species that are close to emerging naturally and do not need their genomes to be artificially manipulated in order to become dangerous.