The only thing stopping AI from being overpoweringly good at audits is the lack of investment in the AI audit space. I’ve trained multiple world-class auditors both directly and indirectly. I also played go competitively and know I have zero chance at doing anything remotely close to world class. Competitive go is harder than audits as the permutation of entry-points and meaningful state-space is far larger and much harder to reason about. AI crushed humans at go in 2016. Someone who put 20-30 million dollars into AI audit R&D would completely flip things. The top of the leaderboard for web2 bug bounties in the United States is an LLM. *Such an AI is not a one-shot LLM but something that can agentically write unit tests and formal verification specs to test its understanding of the codebase. The number of tokens this consumes would be extremely expensive, but likely competitive with what audits cost now. **AI will never completely replace auditors as they never have as much context as a human. But the only thing stopping serious disruption is a lack of serious investment. I would angel invest in such a project, if I saw the team had the right experience with audits and machine learning.