Optimal liability for offensive and defensive AI
How much liability should AI providers bear when their services enable both attack and defence? Liability can improve welfare while increasing harm. Providers sell a common input to productive users, attackers and defenders. Within a defended contest, a higher common price reduces effort without changing attack success or attacker profits, saving resources and improving the […] The post Optimal…
The liability providers bear for AI services that can enable both offensive and defensive actions is a complex issue. While liability can improve overall welfare, it also risks causing more harm if not carefully balanced. AI is sold to both attackers and defenders, so the common price within a defended contest reduces effort without affecting attack success or profits, ultimately saving resources and enhancing the target's security.
However, compensation actually weakens defense and increases attacker profits. To find the optimal liability, it must be balanced against the exclusion of productive uses. Greater competition can lower the optimal liability, but this must end at an outcome that still preserves defense. In situations where cybersecurity access is fixed, a monopoly can justify partial liability but never full liability when provision is still valuable.
When guardrails exist that allow for productive uses, strong competition favors universal guarding as socially beneficial, but it can encourage some providers to remove guarding completely if liability isn't high enough. With a fixed number of providers, enough productive users can lead to a pure equilibrium with universal guarding when liability is high. In such cases, a universal-guarding requirement makes liability unnecessary.
For monopolies, there is a unique liability threshold that determines adoption, while zero liability is optimal when there are enough productive users and the liability parameter falls within a specific range. These insights come from a new paper by Joshua Gans. The findings were first reported on Marginal REVOLUTION.
Written by urgent.news from Marginal Revolution's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
