Research
Study how agents behave in real-world systems and where they fail people, especially in small, rural, and underserved communities.
Agent Accountability Institute
When Agents Control.
When agents control the systems we depend on, someone has to hold them accountable.
The Agent Accountability Institute studies what happens when AI agents take real actions on behalf of people and communities, and publishes open guidelines so those systems stay safe, fair, and answerable to the people they serve.
Why now
Agents are moving from answers to actions.
AI is moving from answering questions to taking actions: routing vehicles, moving money, approving benefits, and scheduling care. Many groups study AI safety in general, and companies are building tools to govern their own agents. A new industry is forming to give agents identities, credit scores, and insurance. No independent institute focuses on public accountability when agents act in the real world. We exist so the public has a voice in the rules those tools follow.
Purpose
People stay in charge.
To make sure that when AI agents control public and personal systems, people stay in charge: they can understand what an agent did, challenge it, and get a human when it matters.
Guidelines
Publish open, practical guidelines that cities, agencies, and operators can adopt.
Testing
Develop simulation and safety tests for agents before they reach the public.
Convening
Bring together communities, operators, researchers, and policymakers to agree on shared standards.
The series
When Agents Control ___
An open research and guidelines series. Each part looks at one domain where agents are starting to act for people, and asks the same questions: Who is accountable? What can go wrong? How does a person get help?
- Part 1in progress
When Agents Control the Fleet
Shared and autonomous vehicle fleets are coming to small and rural communities, often run by AI that dispatches rides and manages vehicles across many owners. This part maps who is involved, tests fleet agents in simulation, and sets guidelines for safety, access for riders with disabilities, and local control. Its adoption path is the Leapfrog Autonomous Mobility Compact, in formation.
- Part 2forthcoming
When Agents Control the Money
Agents are starting to pay bills, move funds, and manage accounts for people. This part looks at consent, spending limits, fraud, mistakes, and how a person reverses an action an agent took in their name.
- Part 3forthcoming
When Agents Control Benefits
Agents are starting to screen, approve, and deny public benefits and subsidies. This part looks at fairness, appeals, explanation, and making sure no one loses support because of an automated error.
- Part 4forthcoming
When Agents Control City Services
From 311 requests to permits and inspections, agents are starting to handle how residents reach their local government. This part looks at transparency, equal service, and the right to reach a human.
- Part 5forthcoming
When Agents Control Care Logistics
Agents are starting to schedule appointments, rides to care, and the delivery of prescriptions. This part looks at missed care, privacy, and what happens when an agent gets it wrong for someone who depends on it.
Use cases
Accountability on both sides.
Agents are starting to act for people. To be accountable, they also need things that people take for granted: an identity, limits, a record, and someone responsible when they get it wrong. Our research looks at both sides.
When agents control your purchases
An agent buys the wrong thing, books the wrong flight, or pays the wrong bill. We study how a person reverses an action taken in their name, who covers the cost, and what protections should exist, much like buyer protection for card payments.
When agents control your identity
Agents can speak in your voice, use your face, and remember what they know about you. We study consent, licensing of a person's likeness, and a person's right to see, correct, and delete what an agent knows about them.
When agents control what you see
When an agent recommends a product, a provider, or a route, it may be influenced by who paid. We study disclosure rules so people know when a recommendation is paid placement.
When agents control who reaches older adults
Fake voices and AI-written messages are already targeting seniors and families. We study warning signs, safeguards, and how agents can protect vulnerable people instead of being used against them.
Principles
What we stand for.
- A human can always be reached.
- People can see what an agent did and why.
- Every action has a responsible party.
- Systems are tested before they are trusted.
- Small and rural communities are designed for, not added later.
About
A research organization in formation.
The Agent Accountability Institute is a research organization in formation, founded by Breanna Faye. It is the home of the “When Agents Control ___” series and the Leapfrog Autonomous Mobility Compact.
About the founder
Breanna Faye is a researcher and designer focused on how emerging technology behaves when it meets real people. For more than 15 years she has built products using accessible design practices, testing how systems work for riders, residents, and users with disabilities, and designing for the people most likely to be left out. An MIT alum who worked with the MIT Media Lab, she brings a futurist and trends research lens to questions of safety and public trust.
Her work in mobility includes advising on Hyperloop transit stations and on planning, software design, and network strategy for connected autonomous vehicle fleets, including work tied to Hyperloop UAE and NEOM. Through her firm, Agentic Futures LLC, she advises public-sector and infrastructure clients on AI strategy, governance, and safe deployment.
She founded the Agent Accountability Institute because AI agents are beginning to take real actions in systems people depend on, and accessibility, safety testing, and human oversight need to be built in from the start, not added after something goes wrong.
Get involved
Help shape the first parts of the series.
We are looking for researchers, cities, transit agencies, operators, and funders who want to help shape the first parts of the series.