Token arms race centralizes AI infrastructure

As advanced models become exploitable via paid API tokens, a competitive dynamic may emerge where attackers pay for abundant model access while defenders must buy costly patches, guardrails, or extra compute to mitigate harm. That pricing asymmetry will favor large institutions that can prepay or vertically integrate defenses, driving further concentration of AI hosting, patching, and security services. — If true, this dynamic would reshape who controls AI capability, raise national‑security and antitrust concerns, and change the focus of regulation from models themselves to economic choke points (compute, tokens, and patch markets).