Accountability

Fiduciary standards for autonomous systems.

Software cannot hold a duty. When an agent allocates, the fiduciary obligations — care, loyalty, prudence — remain exactly where they always were: with the humans and institutions that deployed it. The useful question is what the duty of care requires of a fiduciary who delegates work to a machine.

DUTY OF CARE · DISCLOSURE · OVERSIGHT OBLIGATION · REVISED 2026-07-26

Delegation is not abdication

Fiduciary law has always permitted delegation — to sub-advisers, to index rules, to trading algorithms — under a consistent condition: the fiduciary must select prudently, instruct clearly, and monitor continuously. An autonomous agent is the newest delegate, not a new category of moral actor. Deploying one without the capacity to supervise it is the breach; the agent's sophistication is no defense.

What prudent deployment looks like

Disclosure allocators should expect

Institutions allocating to AI-native strategies should treat the following as baseline diligence: the degree of autonomy in production (research assist vs. constrained execution vs. adaptive behavior); the independence of validation; the constraint register and where limits bind; incident history including every kill-criteria trigger; and the change-control process by which the agent's behavior can evolve. A manager unwilling to answer these has answered them.

The direction of travel

Accountability frameworks are converging on a simple architecture: machines execute; named humans remain answerable; evidence connects the two. The institutions that internalize this early will find that governance, far from slowing agentic adoption, is what makes it insurable, allocatable, and durable — the entire thesis of the QIS ecosystem.

Editorial synthesis of fiduciary principles under ERISA, the Investment Advisers Act, and trust law as applied to delegation and oversight. Not legal advice.