What would it take to trust agent-written crypto code? Five steps and a proof

blaizedsouza · x · 2026-11-21

A discussion on when agent-written cryptographic code can be trusted. @copyconstruct lays out the minimum bar: understand the fundamentals, run experiments exposing flaws in existing designs, quantitatively analyze alternative designs and tradeoffs, run continuous verification loops to prove or disprove the implementation, and communicate all of it clearly. The quoted question pushes harder: test vectors plus a proof, or don't trust it at all.

Original post →

More from coding & agent

coding & agent channel →