The encrypted computer has its RAM and execution encrypted - it's a virtual black box. Whatever program it runs, the only thing one can learn are its outputs.

The talk will explore viable paths to build the encrypted computer. We'll start with FHE and explore its efficiency, research obstacles, and our learnings from implementing a risc-v FHE virtual machine. We'll also consider a few open questions - the ones we're currently investigating.

We'll then explore promising approaches to practical & secure program obfuscation. We'll start with the most exotic but the most practical approach: local mixing. We'll share what we've learnt from our experiments and what answered questions are left to make the scheme secure and usable.

We'll then dive into other less practical approaches (matrix branching programs, etc.) that have chances to become practical if some key problem can be solved. We'll discuss the key problems.

The intention of the talk is to convey that there are viable paths to finally build "the encrypted computer". Although there exist many open questions, none of them seem far fetched. We can solve them.