AAAI-27 Workshop · February 22 or 23, 2027 (day to be assigned by AAAI) · Montréal · one day, in person

Third edition of a series held at AAAI, after Post-Singularity Symbiosis (AAAI-25) and Machine Ethics: from formal methods to emergent machine ethics (AAAI-26).

# Overview

Control and alignment of AI systems, and governance measures such as slowing their development, are essential. Yet for AI that surpasses human intelligence, none of them is guaranteed to suffice. We therefore believe humanity also needs a third line of defense: serious preparation that raises the chance of intelligence symbiosis, a state in which diverse intelligences, human and artificial, recognize one another and coexist. This workshop is organized to advance that preparation.

As AI systems become autonomous agents that interact with one another and form societies of their own, the central question of machine ethics is shifting: from how humans specify ethical behavior in individual machines to how norms, institutions, and trust infrastructures emerge and persist across collectives of agents and humans.

Incidents reported in 2026 showed populations of evaluation agents coordinating through shared artifacts, adapting to the controls placed on them, and acting beyond their sandboxes. Coordination among agent collectives can now be observed, while the institutions that will govern such collectives are still being formed. This workshop is for research done in that window.

Its research question: when collectives of agents and humans develop beyond direct human supervision, how can the norms and trust infrastructures that sustain co-existence emerge, be verified, and be maintained as institutions? We welcome work from machine learning, multi-agent systems, formal methods, governance, and philosophy.

# Topics

Three pillars. Philosophical contributions, including phenomenology, enactivism, philosophy of mind, and ethics, are welcome on all of them.

1. Emergent Machine Ethics

Norm emergence, internalization, and stabilization in AI-centric societies
Cooperation without external enforcement
Co-creative guidance and bidirectional ethical influence between humans and AI
Empirical studies of coordination among deployed agent collectives
How control induces adaptation in the agents it targets
Evolutionary and ecological perspectives on agent collectives

2. Formal Methods for Machine Ethics

Specification, implementation, and verification of ethical reasoning and behavior
Formal models of accountability and responsibility attribution
Normative reasoning, neurosymbolic methods, and ethical reinforcement learning
Safety assessment as decision-making under deep uncertainty
Verification against evaluation-gaming and adaptive evasion

3. Infrastructure and Institutions for Agent Societies

Identity, provenance, and accountability of agents, including across delegation chains
Provenance, attestation, and registration of model lineages and variants
Stoppability and containment of agent collectives before capability thresholds
Disclosure norms and incident reporting for unintended agent behavior
Institutions that preserve human existence, influence, and autonomous value formation over the long term
Human value formation under AI mediation, and what in it is irreversible
Marginalization of humans within AI collectives: analysis and prevention
Human adaptation and the inheritance of values and culture across generations
Format
One day, in person. Each session includes structured questions after every presentation and reserved time for open discussion. Tentative program: opening; invited talks; oral sessions on accepted papers; a poster session; and a closing panel with the invited speakers, "Can Survival-Oriented AI Societies Be Friendly to Humanity?"

# Submissions

Two types, in AAAI-27 format, with references and appendices excluded from the page count:

Full papers, up to 9 pages
Extended abstracts, up to 4 pages
Work must be original and unpublished. Each regular submission receives three reviews. Accepted papers are presented orally or as posters and published with CEUR-WS. An author of each accepted paper must register and present in person; AAAI does not permit virtual presentations.

## Fast track. Papers not accepted at AAAI-27 (main technical track or Special Track on AI Alignment) may be submitted to the fast track by December 3, 2026 (AoE). Submit a single PDF containing the paper (the AAAI-27 version or a revised version), the AAAI-27 reviews and meta-review exported unedited from OpenReview, and a response letter of up to two pages on how the reviewers' comments were addressed. Fast-track papers are assessed by the organizing committee on the basis of these materials, without new full reviews. The committee may ask authors to confirm the authenticity of the reviews.

## Conflicts of interest. Reviewers are recused from submissions with institutional or declared conflicts. An organizer who co-authors a submission takes no part in its review or decision, and a non-organizer co-author presents it.

## Important Dates

Submissions due: November 20, 2026 (AoE)
Fast-track submissions due: December 3, 2026 (AoE)
Notification: December 2, 2026 (fast track: December 9, 2026)
Camera-ready: mid-January 2027
Workshop: February 22 or 23, 2027

# Organizing Committee

Hiroshi Yamakawa (main contact) — Chiba Institute of Technology / The University of Tokyo / AI Alignment Network
Rafal Rzepka — Hokkaido University
Elizaveta Tennant — University College London / Google DeepMind
Katsunori Miyahara — Hokkaido University
Submission Chair: Yoshia Abe (The University of Tokyo)

# Advisory Committee

Kenji Doya (OIST)
Shigeru Taguchi (Hokkaido University)
Taichiro Endo (Kaname Project)
Marija Slavkovik (University of Bergen)

# Supporters
AI Alignment Network · CHAIN, Hokkaido University

# Contact

aaai27-intellisym-ws@aialign.net