Open problem 9
The Responsibility and Liability Gap
- current AI
- frontier AI
- AGI
How should law allocate responsibility among model developers, deployers, users, cloud providers, data suppliers, fine-tuners, auditors, and regulators?
Why it’s foundational
Accountability is impossible if responsibility dissolves into the AI supply chain. Current systems already blur responsibility; more autonomous systems will intensify the problem.
Why it’s difficult
AI harms can be probabilistic, systemic, delayed, or emergent. The causal chain may involve model design, data, interface, prompts, deployment context, user conduct, and organisational negligence. Existing doctrines of product liability, negligence, professional malpractice, and administrative law only partially fit.
Hidden assumptions
A common fiction is that “human oversight” preserves responsibility. It often does not. A nominal human-in-the-loop may be rubber-stamping outputs they cannot understand, contest, or realistically override.
Competing positions
- Strict liability for developers
- Negligence standards
- Deployer liability
- Shared liability
- No-fault compensation funds
- Mandatory insurance
- Criminal liability for reckless deployment
- Regulatory safe harbours for audited systems
What could make progress
Comparative legal analysis; incident databases; litigation studies; experiments on human oversight in real organisations; actuarial models for AI insurance; doctrinal work on causation under probabilistic systems.
What it would change
It would affect insurance, audit incentives, documentation duties, user rights, procurement rules, and compensation systems.
Sub-agenda
- When should developers remain liable for downstream misuse?
- When does human oversight become legally meaningless?
- Should audited compliance reduce liability or increase it?
- Can strict liability work for open-weight models?
- What compensation mechanisms are needed for systemic AI harms?
Priority (editor scoring)
Highly actionable for current AI; less sufficient for AGI-scale risk.
- Importance
- 4/5
- Neglected
- 3/5
- Difficulty
- 4/5
- Actionable
- 5/5
- Robust
- 5/5
- Nat’l+int’l
- 3/5
Where the catalog bears on this
No current catalog instrument resolves this puzzle — which is the point: it is a foundational question the existing rules leave open. Browse the coverage catalog for what the instruments do and don’t say.
Editorial content — a human-authored agenda question, rendered verbatim. No part of this analysis is AI-generated (see the charter).