Descriptions:
At the AI Engineer conference, Jack Cable — co-founder and CEO of Corridor and former senior technical advisor at CISA — delivers a structured talk on what he terms the “AI bug apocalypse”: the convergence of AI-generated code flooding production systems with frontier models’ rapidly improving ability to discover and exploit the vulnerabilities in that code.
Cable grounds the argument in concrete data. Roughly 84% of developers were already using AI coding tools as of last year’s Stack Overflow survey, and 30-40% of companies were actively encouraging their use — numbers he expects have grown significantly since. Simultaneously, frontier models like Anthropic’s Claude (which he references by the Mythos designation in Anthropic’s internal capability charts) are showing rapid improvement on autonomous attack chain benchmarks. The result is a two-sided squeeze: more AI-written code creating larger attack surfaces, and more capable models making exploitation cheaper and faster for adversaries.
The core of Cable’s argument is that reactive vulnerability patching is the wrong response to this dynamic. He advocates instead for structural hardening, using memory-safe languages like Rust as a primary example. Citing Google data on the Android operating system, he notes that simply writing new code in memory-safe languages dropped the share of memory safety vulnerabilities from roughly 75% in 2019 to around 30% in 2022 — without rewriting existing code. His policy recommendation: invest now in one-time rewrites of critical open-source libraries, because programmatic language guarantees remain effective even as model attack capabilities continue to improve.
📺 Source: AI Engineer · Published July 12, 2026
🏷️ Format: Deep Dive
