A first map
of the conversation.
An inaugural synthesis of expert perspectives, areas of agreement, unresolved tensions, and questions for further research.
50 Voices initiativeAI SAFETY MODELS / AN AGENTICERT INITIATIVE
Convening a broader
view of AI safety.
Safety spans more than a model’s behavior. Agenticert is bringing perspectives together to explore how design, evidence, oversight, and real-world outcomes connect.
THE PROPOSED FOUNDING INVITATION
Voices.
Voices.
One evolving conversation.
A proposed invitation to experts, researchers, leaders, and practitioners. A range of perspectives, with room for disagreement.
An invitation goal—not a confirmed participant count.LEADERSHIP THROUGH CONVENING
Better questions.
A more complete picture.
AI Safety Models is an emerging publishing and collaborative research initiative convened by Agenticert. It creates space to connect technical, organizational, and societal perspectives on safer AI.
The starting idea is an evolving framework that can be challenged and improved. It is an inquiry into safety—not a certification standard or a finished answer.
Read the initiative’s proposalA STARTING POINT TO CHALLENGE
Seven layers.
Open to new thinking.
Explore the proposed safety spectrum. Contributors are invited to question the boundaries, add missing dimensions, and reshape the framework.
From the intent of a system to its effects in the world.
01Purpose & values
Who should the system serve?
Begin with the goals, rights, and acceptable trade-offs that shape an AI system. Make the people affected by those decisions part of the discussion.
02Data & training
What shapes its foundations?
Examine where data comes from, how it is handled, and which training choices influence the system’s capabilities and risks.
03Model behavior
How does it act—and fail?
Describe capabilities, limitations, and failure patterns in the context where the system will be used.
04Evaluation & security
What evidence supports confidence?
Consider testing, adversarial challenges, safeguards, and the boundaries of what an evaluation can establish.
05Human oversight
Who can question or intervene?
Make decision ownership, escalation, and meaningful human intervention visible.
06Deployment & operations
What changes in the real world?
Connect deployment conditions, access, incident response, and ongoing review to the safety conversation.
07Societal & systemic outcomes
What happens beyond the product?
Consider effects on people, institutions, information, and the wider systems in which AI operates.
RESEARCH THAT STAYS OPEN
Publish. Discuss. Refine.
The proposed publications are a way to bring perspectives into public view and keep developing the questions.
A first map
A continuing
exchange of ideas.
A publication planned every two months, exploring frameworks, practical lessons, evidence, and thoughtful debate.
AI Safety Models MagazineEditorial independence is central to the proposal. Sponsorship supports the work; it does not purchase conclusions or contributor selection. Publications and participation remain subject to development and review.
ONE MISSION, DIFFERENT CONTRIBUTIONS
Broader thinking.
Practical visibility.
Our research initiative and flagship products share a commitment to making complex AI systems easier to understand.
FLAGSHIP PRODUCT
Agenticert Flow
Explore the infrastructure, dependencies, and data paths behind an AI application.
FLAGSHIP PRODUCT
AI Transparency
Give people an accessible explanation of AI systems and the evidence behind a disclosure.
The initiative informs the wider conversation. It does not certify these products or establish a safety standard for their users.
BRING YOUR PERSPECTIVE
Help shape what comes next.
Contribute expertise, explore a research partnership, or discuss support for the proposed publications.