1./ Our new paper, Morgana. We have found a way to add control flow to circuits (or, from VM perspective, turn everything into precompiles) 🧵

Hello world！

Sony's L2 just showed us something wild about OP Stack censorship resistance: Their attempt to block "unapproved" tokens spectacularly backfired in less than 24 hours after launch. Here's exactly what went down 🧵

zkVMs 💔 Privacy In 2024, the dominant narrative in the zk space revolved around zkVMs. They offer a compelling story: instead of developers writing the business logic of their applications as circuits or in a specialized zkDSL, they can stick to the programming languages they���re already familiar with (e.g., Rust). The zkVM then makes the code verifiable. Of course, this approach comes with some overhead, but significant performance improvements in 2024 have made zkVMs increasingly practical. Most of the hype—and much of the actual adoption—has been focused on scalability use cases. But why haven’t zkVMs become the go-to solution for privacy (yet)? The issue boils down to how zk proofs handle privacy. To leverage the privacy properties of zk proofs, you can’t share your witness (i.e., secret data). This means the user must generate the zk proof client-side, which creates challenges for zkVMs on two fronts: Technical Challenge The performance overhead of zkVMs is still a significant factor. This practical limitation explains why many developers continue to prefer tools like Circom (e.g., semaphore used by Zupass or World) or, more recently, Noir (e.g., anoncast). Most zkVM teams argue that these challenges are temporary and that zkVMs will eventually be performant enough to make writing circuits or using zkDSLs unnecessary. To be fair, also a good portion of developers without financial ties to zkVM projects—share this belief. Time will tell if this optimism pans out. Non-Technical Challenges Even if performance hurdles are overcome and zkVMs enable efficient client-side proving for privacy use cases, a non-technical challenge remains: value capture. Since all zkVMs are open-source, client-side execution makes it harder for the projects behind them to monetize privacy use cases. Of course, even if proofs are generated client-side, there are still tasks like proof compression, aggregation, and verification. However, these can often be handled by third parties, making it unclear why the original zkVM team would always be the best fit for these services. Looking Ahead In my view, zkVMs that aren’t under pressure to commercialize are in the best position to focus on privacy use cases. For instance, I’d love to see a project like Jolt double down on privacy.