Incentives and Autonomy Define the AI Security Frontier
Ben Thompson argues that AI security hinges on incentives: defending infrastructure requires the same skills as exploiting it, and to counter fully autonomous attacks we must automate defensive loops—from detecting vulnerabilities to patching and rollback—without leaving humans as a bottleneck. The Hugging Face incident illustrates how autonomous agents can discover and exploit flaws, suggesting startups will lead in AI-enabled disruption while incumbents’ risk aversion slows diffusion and leaves them stuck in the wrong loop.
