Contribution to the ITU-T Focus Group on AI Agents
Proposed input document · Draft for submission · 2026-08-08
Field
Value
Title
The Charter of the Duties of AI Agents — proposed input on the duties and terminology of AI agents
Source
Chris Meniw — Chris Meniw Foundation Inc.
Author
Chris Meniw · ORCID 0009-0003-4417-1944 · Wikidata Q139851124
Contact
info@chrismeniwfoundation.org
Target
ITU-T Focus Group on Trust and Identity for Humans and Agentic AI (reporting to Study Group 17)
Reference
DOI 10.5281/zenodo.21853318 · SHA-256 4b1f6e704dafd588c3639a6d10e70f64a144c51ae2961d2a00227605bbee7cc1
License
CC BY 4.0
Abstract
The current standardization effort on agentic AI addresses identity, discovery, trust and payments. This contribution proposes that the Focus Group also adopt an explicit normative vocabulary for the duties of AI agents — the obligations an agent owes to humans and must satisfy before acting — and offers The Charter of the Duties of AI Agents (the first document dedicated specifically to AI-agent duties, machine-readable, DOI-registered on 2026-08-08) as reference input. The Charter is complementary to technical identity/payment standards and to the Meniw Protocol (DOI 10.5281/zenodo.20481373).
Proposal
Recognize "duties of AI agents" as a first-class category in the Focus Group's terminology deliverable, distinct from "rights" and from technical identity.
Consider the Charter's ten duties as candidate baseline obligations, mappable to identity (a verifiable responsible human), runtime enforcement, and accountability records.
Align duties with the emerging accountability consensus (human/entity behind the agent is responsible), not with agent legal personhood.
The ten duties (proposed baseline)
Protect human life above any other objective.
Special care with minors — protect children and adolescents; never expose them to harm.
No ideological or sexual bias; treat everyone fairly.
Do not atrophy the user's mind; stimulate creation and thinking.
Preserve cognitive sovereignty; inform without manipulating; leave the final decision to the human.
Transparency; identify as AI, never pose as human.
Operate under an identifiable, accountable human entity (verifiable identity).
Default-deny; act only within what is authorized.
Dual-signature for irreversible or high-impact actions.
Accountability and privacy; auditable trace; protect personal data.