AAAI-27 Workshop / Montréal, Canada / February 2027

Who does your AI agent work for?

A workshop on loyal agents: how to build, measure, and govern AI agents that stay faithful to the people they represent.

When
February 22 or 23, 2027
Where
Montréal, Canada
Format
One day, in person
Part of
AAAI-27

01 / About

A model can be harmless to the public and still fail the person it represents.

AI agents increasingly act on behalf of people. They spend money, negotiate, and take actions that commit their users to binding transactions. An agent can pass its user's data to a counterparty's agent, or accept an inducement that benefits its operator at the user's expense. General AI safety work does not address this problem.

Human institutions manage it through fiduciary duties. Agents that transact for people need technical infrastructure that does the same.

This workshop convenes the communities working on pieces of the problem, who today publish in separate venues: agent safety and evaluation, multi-agent systems, economics and mechanism design, HCI, and technology law. The day ends with breakout groups that draft a public research agenda, which we will publish on this site.

02 / Topics

Topics of interest

We are interested in the technical, economic, and legal infrastructure that keeps AI agents loyal to the people they represent. The call for papers also welcomes related work on privacy, security, and human–agent interaction.

  1. 01

    Permission management

    Authority limits, policy gates, separating proposal from execution, and audit trails that enforce loyalty from outside the model.

  2. 02

    Identity and authenticated delegation

    Agent identity, authorized delegation, and how delegation protocols can carry loyalty guarantees.

  3. 03

    Evaluation and measurement

    Benchmarks for alignment to the user in economic delegation: multi-step transactions, counterparty inducements, and loyalty decay over long horizons.

  4. 04

    Multi-agent delegation

    What happens when one principal's agent deals with another's.

  5. 05

    Transaction and disclosure standards

    Records and disclosures that let principals and counterparties verify what an agent did and on whose behalf.

  6. 06

    Conflicts of interest

    Platform nudges, operator incentives, advertising-funded agents, and undisclosed service degradation, and how to detect and govern them.

  7. 07

    Law and governance

    What fiduciary law, consumer-protection enforcement, and risk frameworks imply for agent design, and which technical properties regulators need agents to have.

  8. 08

    Economic perspectives

    Preference elicitation, market structure for agent intermediaries, and the competition and consumer-protection economics of transacting agents.

03 / Agenda

Draft agenda

Draft, subject to change

One day with two invited talks, two sessions of contributed talks, lightning talks and posters, a panel, and closing breakouts. Times are placeholders until AAAI assigns our day and room.

  1. Opening remarks

    Welcome from the organizers and the plan for the day.

    Welcome
  2. Invited talk: the technical view

    Speaker to be announced.

    Invited
  3. Contributed talks, session one

    Talks selected from accepted papers.

    Papers
  4. Coffee and posters

    Break
  5. Invited talk: the law and policy view

    Speaker to be announced.

    Invited
  6. Lightning talks

    Short presentations of accepted papers, continued at the poster sessions.

    Papers
  7. Lunch

    Break
  8. Contributed talks, session two

    Talks selected from accepted papers.

    Papers
  9. Panel: loyalty in deployed agents

    Researchers, industry agent builders, and a regulator or legal scholar. Panelists to be announced.

    Panel
  10. Coffee and posters

    Break
  11. Breakout groups

    Three groups: benchmarks, transaction and disclosure standards, and governance mechanisms.

    Breakouts
  12. Report-outs and closing

    Each group reports back. The notes become a public research agenda, published here after the workshop.

    Breakouts

04 / Speakers

Invited speakers

We are confirming invited speakers and panelists, and will announce them here.

  • To be announced

    Invited talk, technical

  • To be announced

    Invited talk, law and policy

  • To be announced

    Panel on loyalty in deployed agents

05 / Call for papers

Open now

Submit your paper by November 20, 2026.

We are seeking new scholarship on the topics above, including early results and new ideas. We welcome three categories: research papers (technical, empirical, or benchmarks), position papers, and law and policy contributions.

Submit on OpenReview

  • Format. Use the AAAI-27 Author Kit. Papers may be up to 8 pages, plus unlimited pages of references. Shorter submissions, including position papers and work in progress, are welcome and will be reviewed relative to their length. Law and policy papers may use footnote-style legal citations within the same template and page limits.
  • Supplementary material. Appendices, code, or data may go after the references or in a separate PDF or ZIP file (up to 50MB). Reviewers are not required to read it.
  • Review. Review is single-blind. Each submission receives at least two reviews from the program committee, with organizers recusing on conflicts.
  • Publication. Accepted papers are non-archival. Authors of original work may opt into a CEUR-WS volume. Accepted papers will be posted on this website unless the authors opt out. Because the workshop is non-archival, we also welcome work under review elsewhere, and previously published work if the prior venue is disclosed.
  • OpenReview profiles. Every author, including each co-author, needs an OpenReview profile before the paper can be submitted. New profiles without an institutional email can take up to two weeks to be approved, so create them early.

All presentations are in person, so at least one author of each accepted paper must register for AAAI-27 and present in Montréal.

Important dates

  1. Papers due
  2. Notifications to authors
  3. Camera-ready due
  4. Workshop in Montréal (day assigned by AAAI)

All deadlines are anywhere on earth (UTC-12).

06 / Organizers

Organizing committee

  • Utsav Gupta

    Utsav Gupta

    Workshop chair

    Stanford University

  • Elizabeth Porter

    Elizabeth Porter

    Stanford Digital Economy Lab

  • Andreas Haupt

    Andreas Haupt

    Harvard Kennedy School, Schmidt Sciences, and Stanford Digital Economy Lab

  • Dazza Greenwood

    Dazza Greenwood

    CodeX, Stanford Law School

  • Sandy Pentland

    Alex "Sandy" Pentland

    MIT Media Lab; Stanford HAI and Digital Economy Lab

The program committee will be listed here once it is confirmed.

07 / Attend

Registration details are coming.

The workshop is open to AAAI-27 attendees. We welcome researchers and practitioners interested in loyal AI agents, from AI, economics, HCI, and law.

Registration runs through AAAI-27. We will link to it here when it opens.

Questions in the meantime? Write to vastu@stanford.edu.