I'd love to see software adopt strong capabilities-based models that enforce boundaries even within parts of a program. That is, with the principle of least authority (POLA), code that you call is passed only the capabilities you wish (e.g. opening a file, or a network socket), and not everything that the current process has access to. Thomas Leonard's post (https://roscidus.com/blog/blog/2023/04/26/lambda-capabilitie...) covers this in great detail, and OCaml's newer Eio effect system will has aspects of this too.
The Emily language (locked-down subset of OCaml) was also interesting for actively removing parts of the standard library to get rid of the escape hatches that would enable bypassing the controls.
Linux has seccomp, but I think that was changing the access for an entire process. The language-focused aspect seems useful to me, from that application aspect where maybe I want access to something, but I don't want to pass that access on to all the code that I might call from a library.
The Emily language (locked-down subset of OCaml) was also interesting for actively removing parts of the standard library to get rid of the escape hatches that would enable bypassing the controls.