While containers have some useful properties, it was never intended to be, and never really functioned as a strict security boundary. We've duct-taped around that, and it's reasonably good now, but that only goes so far.
I typically say that containers (and any other isolation that shares a kernel) are good for "mostly trusted" workloads, like different teams at the same company. You want isolation against accidents more than intentional attacks.
VMs are good for just about everything if you are careful (for example what devices and hardware are exposed) but if you want ultimate isolation you want completely separate hardware. It is the only way to be sure against hardware bugs and side-channels or VM bugs.
Shrink your attack surface.
Use a completely locked down seccomp. Use nsjail or gVisor for containers. Use microvm or libkrun for full OS.
Lesser attack surface is what matters. Virtualization is only half of the story.
What does that even mean? Bugs are finite, at some point they'll just all be patched and you'll be left with bug-free software.
AI bug discovery getting better is an incredible advantage for defenders.
From "The Tao of programming": "The highest sounds are hardest to hear. Going forward is a way to retreat. Great talent shows itself late in life. Even a perfect program still has bugs. "