The 'Bugpocalypse': LLMs Are Unearthing Long-Hidden Software Vulnerabilities at Furious Pace

esrtweet · x · 2026-10-24

The author cites the so-called "Bugpocalypse": LLMs are being used at a furious rate to find long-hidden cracks in software infrastructure, with one claim that 3% of all compromises in computing history happened in the last week — which the author believes.
Responding to calls to lock out or heavily guardrail public access to frontier models, the author argues that when lockpicks are outlawed, only outlaws will have lockpicks — and Chinese open-weight models will make such tools ubiquitous regardless.
Two futures: one where only bad actors can find undiscovered vulnerabilities, another where everyone aggressively red-teams their own software so holes get fixed before exploitation. The corpus of human code is huge but finite, so its defects are finite too — and we now have "a million friendly robots" to fix them. "The only way out is through."

More from AGI Musings