#dusk $DUSK @Dusk
Running proof generation locally to benchmark circuit performance, I noticed something that made me go back and read the cryptography team's own writeups instead of the marketing pages.
PLONK's actual numbers are what make the compliance case work, not just the privacy angle. Verification time stays around 6-9 milliseconds regardless of circuit size — proving time scales with circuit complexity (roughly 5.46 seconds for a 2^16-gate circuit on modest hardware), but the verifier's side stays fast and constant. That asymmetry matters more for regulated finance than people give it credit for: an auditor or counterparty checking a proof isn't burning meaningful compute every time, even as the underlying transaction logic gets more complex.
What I hadn't expected to find was that PLONK itself had a real disclosed vulnerability, not just theoretical risk. Dusk's research team found a critical issue in how the Fiat-Shamir transformation was implemented — the piece that turns an interactive proof into a non-interactive one by hashing challenges instead of a live verifier sending them. The original implementation didn't hash the public inputs early enough, which weakened the soundness guarantee. Trail of Bits coordinated the disclosure, Dusk patched it before mainnet, and pushed the fix publicly rather than sitting on it.
That's the detail I keep sitting with — a compliance-focused chain built on a cryptographic proof system that had an actual soundness bug in production-adjacent code, caught and fixed before it mattered. I don't know how many other implementations using PLONK elsewhere were still vulnerable when this became public, or how long the gap was between disclosure and other projects patching their own forks.
Running proof generation locally to benchmark circuit performance, I noticed something that made me go back and read the cryptography team's own writeups instead of the marketing pages.
PLONK's actual numbers are what make the compliance case work, not just the privacy angle. Verification time stays around 6-9 milliseconds regardless of circuit size — proving time scales with circuit complexity (roughly 5.46 seconds for a 2^16-gate circuit on modest hardware), but the verifier's side stays fast and constant. That asymmetry matters more for regulated finance than people give it credit for: an auditor or counterparty checking a proof isn't burning meaningful compute every time, even as the underlying transaction logic gets more complex.
What I hadn't expected to find was that PLONK itself had a real disclosed vulnerability, not just theoretical risk. Dusk's research team found a critical issue in how the Fiat-Shamir transformation was implemented — the piece that turns an interactive proof into a non-interactive one by hashing challenges instead of a live verifier sending them. The original implementation didn't hash the public inputs early enough, which weakened the soundness guarantee. Trail of Bits coordinated the disclosure, Dusk patched it before mainnet, and pushed the fix publicly rather than sitting on it.
That's the detail I keep sitting with — a compliance-focused chain built on a cryptographic proof system that had an actual soundness bug in production-adjacent code, caught and fixed before it mattered. I don't know how many other implementations using PLONK elsewhere were still vulnerable when this became public, or how long the gap was between disclosure and other projects patching their own forks.