Thanks to Arantxa Zapico, Benedikt Wagner, and Dmitry Khovratovich from the EF cryptography team for their contributions, and to Ladislaus, Kev, Alex, and Marius for the careful review and feedback.
The zkEVM ecosystem has been sprinting for a year. And it worked! We crossed the finish line for real-time proving!
Now comes the next phase: building something mainnet-grade.
From speed to security
In July, we published a north-star definition for realtime proving. Nine months later, the ecosystem crushed it: proving latency dropped from 16 minutes to 16 seconds, costs collapsed 45×, and zkVMs now prove 99% of all Ethereum blocks in under 10 seconds on target hardware.
While the major performance bottlenecks have been cleared by the zkEVM teams, security still remains the elephant in the room.
The case for 128-bit provable security
Many STARK-based zkEVMs today rely on unproven mathematical conjectures to hit their security targets. Over the past months, STARK security has been going through a lot, with foundational conjectures getting mathematically disproven by researchers. Each conjecture that falls takes bits of security with it: what was advertised as 100 bits might actually be 80.
The only reasonable path forward is provable security, and 128 bits remains the target. It’s the security level recommended by standardization bodies and validated by real-world computational milestones.
For zkEVMs, this isn’t academic. A soundness issue is not like other security issues. If an attacker can forge a proof, they can forge anything: mint tokens from nothing, rewrite state, steal funds. For an L1 zkEVM securing hundreds of billions of dollars, the security margin is not negotiable.
Three Milestones
For us, security and proof size are both critical—but they’re also in tension. More security typically means larger proofs, and proofs must stay small enough to propagate across Ethereum’s P2P network reliably and in time.
We are setting three milestones:
Milestone 1: soundcalc integration Deadline: End of February 2026
To measure security consistently, we created soundcalc: a tool that estimates zkVM security based on the latest cryptographic security bounds and proof system parameters. It’s a living tool and we plan to keep integrating the latest research and known attacks.
By this deadline, participating zkEVM teams should have their proof system components and all of their circuits integrated with soundcalc. This gives us a common ground for the security assessments that follow. (For reference, see examples of previous integrations: #1, #2)
Milestone 2: Glamsterdam Deadline: End of May 2026
100-bit provable security (as estimated by soundcalc)Final proof size ≤ 600 KiBCompact description of recursion architecture and sketch of its soundness
Milestone 3: H-star Deadline: End of 2026
128-bit provable security (as estimated by soundcalc)Final proof size ≤ 300 KiBFormal security argument for the soundness of the recursion architecture
Recent cryptographic and engineering advances make hitting the above milestones tractable: compact polynomial commitment schemes like WHIR, techniques like JaggedPCS, a bit of grinding, and a well-structured recursion topology can all contribute to a viable path forward.
Recursion is particularly worth highlighting. Modern zkEVMs involve many circuits composed with recursion in custom ways, with lots of glue in between. Each team does it differently. Documenting this architecture and its soundness is essential for the security of the entire system.
The path forward
There’s a strategic reason to lock in on zkEVM security now.
Securing a moving target is hard. Once teams have hit these targets and zkVM architectures stabilize, the formal verification work we’ve been investing in can reach its full potential. By H-star, we hope the proof system layer will have mostly settled. Not frozen forever, but stable enough to formally verify critical components, finalize security proofs, and write specifications that match deployed code.
This is the foundation that is required to get to secure L1 zkEVMs.
Building foundations
A year ago, the question was whether zkEVMs could prove fast enough. That question is answered. The new question is whether they can prove soundly enough. We are confident they can.
On our end:
In January, we’ll publish a post clarifying and formalizing the milestones above.We will follow up with a technical post outlining proof system techniques for reaching the security and proof size targets.At the same time, we will be updating Ethproofs to reflect this shift: highlighting security alongside performance.We are here to help throughout this process. Reach out to the EF cryptography team.
The performance sprint is over. Now let’s strengthen the foundations.
