Sandbox alone isn't sufficient because when loading a malicious extension into a session you might be exposing your environment variables to it. You also need explicit permissions, so that only relevant extensions can access secrets like this.
And even if this exists, and you are sandboxed, with all guardrails and such, if you trust an extension that later gets compromised, you'd get the same sadness. It's the problems with trusting trust all over again...
Maybe sandboxing, plus freezing updates for at least a week, then on each update reviewing the permissions for each extension, then reviewing the extension itself for compromises. Something that should be done not by the user but by the supply chain itself.
