Going Dark, and the era of law enforcement hacking
Points and comments are a snapshot, not live.
AI-powered bug hunting could eliminate remotely-exploitable vulnerabilities, driving law enforcement to demand backdoors.
The author argues that AI models like Anthropic's Mythos are finding software vulnerabilities so efficiently that major software may soon run out of remotely-exploitable bugs. This would end law enforcement's reliance on purchased hacking tools (e.g., GrayKey, Pegasus) and restart demands for 'exceptional access' backdoors. Such pressure could lead to intentional backdoors in US systems, making them vulnerable to foreign adversaries like Salt Typhoon. The author fears a slide toward nationalized, backdoored software in many countries, comparing the dynamic to the post-2014 'Going Dark' debate but with higher stakes.
What commenters are saying
Commenters split into three camps. Some welcome reduced government hacking capability, calling it 'better than the current situation.' Others argue governments will simply shift to legal pressure for backdoors, making security worse for everyone. A third camp doubts the premise, noting the gap between elite bug-finding AI and everyday software's persistent vulnerabilities, and questions whether self-hosting or open-weight models offer real protection against poisoned training data or government mandates.