Everything is about to “go dark”
Matthew Green analyzes how frontier AI models (Anthropic’s Mythos, OpenAI, Chinese labs) are accelerating vulnerability discovery at scale, forcing defenders into a rapid patching cycle. The article traces how law enforcement historically bypassed encryption via purchased zero-days and exploit tools when cryptography blocked access; as AI systematically exhausts the supply of remotely-exploitable bugs in well-maintained software, law enforcement faces a genuine “going dark” scenario where their workaround vanishes. Green argues this paradoxical outcome—where better AI security creates a surveillance capability gap—will likely trigger renewed policy pressure for exceptional access mechanisms, fundamentally shifting the cryptography and security architecture debates.