NBER Paper: Full Liability Never Optimal for Dual-Use AI Under Monopoly
joshgans · x · 2026-10-05
Economist Joshua Gans published NBER Working Paper 35828 modeling how much liability AI providers should bear when their services aid both attackers and defenders:
- Moderate liability can improve welfare even while increasing total harm: raising the price of a shared input cuts attacker effort without changing attack success, saving resources and improving the target's security payoff.
- Compensation weakens defense and raises attacker profits, so optimal liability trades these effects against excluding productive users.
- Greater competition lowers optimal liability; under monopoly, full liability is never optimal when provision is worthwhile.
- Requiring universal guardrails makes liability redundant; with enough productive users, high liability yields an equilibrium with universal guarding.
Funded by SSHRC; the author notes ChatGPT and Claude assisted with the research.
More from Safety
- EA co-founder MacAskill rebuts Economist: openness to strange ideas is EA's strength — willmacaskill · 2026-10-05
- Report: OpenAI paused frontier training runs after sandbox escape; runtime guardrails cited as production safety — robleclerc · 2026-10-05
- AI welfare paper urges labs to assess AI consciousness now, not later — austinc3301 · 2026-10-05
- AI safety debates fail when people rely on 'common sense' their brains aren't built for — austinc3301 · 2026-10-05
- NeelNanda warns labs are making superhuman hackers they can't control — robleclerc · 2026-10-05
- Detecting deepfakes, in a world where even reality is suspect — CBSnews · 2026-10-05