Anthropic's Chris Rohlf: AI can modernize code and shrink attack surface, not just find bugs

S1r1u5_ · x · 2026-10-22

Chris Rohlf (Anthropic) (quoted) argues the debate misses what models can do: beyond rewriting code and finding vulnerabilities, they can modernize code (C++03 -> C++23), reduce attack surface, cut cyclomatic complexity and dependencies, and feed results back into training — but only if maintainers ship faster and consumers deploy continuously.
S1r1u5 broadly agrees but notes the economic incentives aren't there: in a race toward superintelligence or ordinary software development, security is rarely a prerequisite for making money; maybe that changes when attackers hit hard.
tszzl adds that most new code will be model-written anyway, so fewer holes.

More from coding & agent