Part 7/13:
AI codebases are written and maintained by humans.
While future developments in AI coding might reduce reliance on human oversight, current control relies on security protocols, updates, and kill switches.
Remote and out-of-band management allows for the shutdown of systems regardless of active control.
The Not-So-Serious Concerns with Existing Control Layers
Shapiro emphasizes that multi-layered defense-in-depth strategies—much like cybersecurity best practices—are sufficient for managing current AI risks. These include:
Physical security (power, hardware access)
Cybersecurity (firewalls, intrusion detection)
Administrative policies (training, procedures)