Security of Enclaves & Consensus

According to most people in the industry and academics alike, it is inadvisable based on the number of reported exploits of TEEs to consider tying the security of a blockchain to the trust guarantees of the hardware. In large measure we agree that all other projects should take the wisdom of this thinking. But hear the logic for the argument for the exceptional circumstance we present.

If the outputs from a randomly selected(using a VRF) SEV and SGX are the same we can be assured of the executional integrity and create a proof of correct execution. An attacker would theoretically need to find vulnerabilities within each of these systems, run both as nodes, be selected at the same time to match the outputs, for them to then produce a falsifiable computation. The likelihood of such an undertaking being something any individual, group, state actor could accomplish is what we’re looking at.

The likely event we consider are outcomes related to safety, wherein a compromised machine which produces outputs modifying results different from the inputs will not match the results from the machine of another manufactured TEE.

This would simply result in those executions not being given to consensus validators. If the issue were widespread in as much as there were many compute nodes mounting an attack using one of the hardware the most of what would be accomplished is there would be a stall in the block production.

We can look forward to scenarios where we are using 3 different machines so that even if one of them becomes malicious the 2 others will guarantee that block production does not stall while the community works together to resolve issues on the compromised one.