Pushback on defense-dominance: attackers need one loose brick, defenders guard the whole wall
___Atin___ · x · 2026-10-26
Atin responds to voooooogel's software-simulation defense argument: it only holds if defender models and harnesses match attacker capability. One model generation can run a million costly simulations, only for a more capable model to puncture them — defenders guard the whole wall, attackers need one loose brick.
He agrees formal verification is the direction but sees it as distant: currently 0% of his software stack is verified, and he doesn't expect that to change quickly.
Related event: Debate: Is AI-Era Software Security Defender's or Attacker's Game?(3 posts)→