Safe delegation of AI actions through collective review and voting rules
Delegating Authorization to Misaligned Agents: Coalitional Alignment and Safe Control
Computer Science and Game TheoryArtificial IntelligenceMachine Learning
Summary
When AI systems make decisions over time, it’s hard to keep them safe if they don’t perfectly follow human goals. Having humans approve every action is too slow, so the authors study letting other AI agents review decisions. They find a way to ensure overall safety by requiring a group of reviewers to agree in a specific way, even if some reviewers disagree or are imperfect. Their method guarantees the AI system’s decisions won’t be worse than a baseline and works even when reviewers try to act strategically.
What this means in practice
- •For ai system developers: Design AI systems that delegate safety-critical decisions by combining multiple reviewer agents’ votes to maintain safety guarantees.
- •For autonomous system operators: Implement decentralized approval mechanisms for autonomous agents to reduce reliance on human oversight while ensuring no degradation from baseline policies.
Authors
Natalie Collina, Surbhi Goel, Aaron Roth, Sikata Bela Sengupta
Abstract
Long-running AI agents create a control problem: each action they take changes the state, which in turn affects the trajectory of future actions. If the agent is not fully aligned, then guaranteeing safety requires approving consequential actions before allowing them to be executed. But requiring human approval at every step makes attention a bottleneck. Delegating review to other AI agents raises the same alignment problem: the reviewers may themselves be misaligned. We identify a condition on a reviewing panel that is weaker than individual alignment yet necessary and sufficient for a guarantee that the principal fares at least as well in expectation as under a designated baseline policy. Each reviewer agent reports whether an action proposal made by a proposer agent improves its own utility relative to the baseline. We show that a threshold rule tolerating $k$ disapprovals is safe exactly when, after any $k$ reviewers are removed, the principal's utility can be written as a nonnegative combination of the remaining reviewers' utilities, plus a term that is nonnegative on every feasible proposal. We call this property $k$-robust coalitional alignment. The characterization lifts to sequential control: in a discounted MDP with an arbitrary proposer agent, safety at every state is both necessary and sufficient for the induced policy to match or improve on the baseline. When reviewers vote strategically, full-panel coverage in reward-function space guarantees that every Nash equilibrium is safe under the unanimous approval rule; in contrast, more permissive thresholds can admit unsafe equilibria even when reviewers are individually aligned. Experiments with existing reviewer models show that collective review can remain sound without an aligned individual, even when some disapprovals are tolerated.