Likely AI-generated. Pangram flags 100% of this text as AI-written.
1
CAAF: Making Structural Gaps in AI Safety Visible and Measurable
ProposalGrant
Closes October 1st, 2026$0raised
$5,000minimum funding
$20,000funding goal
Offer to donate
37 daysleft to contribute
You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
Project summary
Current AI safety frameworks miss structural risks by design — in what counts as safe, how it's measured, and whose input shapes the system. I am building CAAF (Cooperative Alignment Assessment Framework), a four-dimensional audit methodology to make these gaps visible and measurable. This funding supports three months of full-time work on a peer-reviewed position paper building on existing published research.
What are this project's goals? How will you achieve them?
Develop a peer-reviewed position paper building on "Whose Safety? Linguistic Hegemony as Designed Fragility" (BTC IV, SSRN) that argues why current AI safety frameworks systematically miss structural risks — not only in language, but in what counts as safe behavior, how performance is measured, and whose input shapes design decisions.
The paper establishes the theoretical foundation of the Cooperative Alignment Assessment Framework (CAAF), a four-dimensional audit methodology:
Behavioral Accountability (D1)
Cultural and Linguistic Integrity (D2)
Measurement Inclusivity (D3)
Stakeholder Participation Depth (D4)
BTC IV is already published. This is refining and strengthening existing work for peer-reviewed submission (target: AI & Society, Ethics and Information Technology, or equivalent), not starting from zero. In parallel, I am co-authoring two papers with Prof. Akira Tokuhiro (Ontario Tech University): one on AI-driven structural capability loss, and one on integrating CAAF with the LENDIT measurement framework for audit protocol design.
How will this funding be used?
Living expenses to enable full-time research. I am an independent researcher with no institutional affiliation or salary. Without funding, research competes directly with finding consulting income, slowing the position paper from 3 months to 6–9 months. This matters because AI governance standards are being institutionalized now — EU AI Act implementation, AISI evaluation protocols, frontier model safety cases. Dimensions built in during this formative period shape what gets measured for the next decade.
Without funding, research competes directly with finding consulting income, slowing the position paper from 3 months to 6–9 months.
Who is on your team? What's your track record on similar projects?
Tomoko Mitsuoka — independent AI governance researcher, Chiba/Tokyo, Japan. Native Japanese speaker, fluent English. MA in European Politics and Democracy Studies (University of Kent). 25+ years of qualitative research (700+ stakeholder interviews across pharmaceutical, healthcare, and technology sectors). Appointed Expert (Japan), Global AI Ethics Institute, Paris.
Track record in this field (2026):
Published four working papers (Beyond Technical Compliance series, SSRN)
Presented at LSAIR 2026, KU Leuven
Four research contributions with Apart Research (CBAAC, Overseer Manipulation, Embedded Loyalty, AI Welfare)
Advanced to MATS Autumn 2026 Stage 3 (top ~5% of 3,500+ applicants)
Collaborator: Prof. Akira Tokuhiro (Ontario Tech University), nuclear engineer and applied mathematician, developer of the LENDIT measurement framework. Co-authoring two papers.
What are the most likely causes and outcomes if this project fails?
Most likely cause of failure: the position paper is not accepted at a peer-reviewed venue on first submission. This is a normal part of academic publishing — the paper exists regardless and can be revised and resubmitted. The theoretical work and CAAF framework do not depend on any single publication outcome.
A deeper failure would be if CAAF's four dimensions prove impossible to operationalize into measurement protocols. This is a genuine risk, and it is why the current phase focuses on theoretical foundation rather than promising a working audit system. The co-authored work with Prof. Tokuhiro exploring LENDIT as a measurement architecture is specifically designed to test this question. Even a negative result — discovering which dimensions resist quantification — would be a useful contribution to the field.
The work would not be abandoned in either scenario. It would continue at reduced pace.
How much money have you raised in the last 12 months, and from where?
$0. This work has been entirely self-funded with no existing grants, institutional support, or external funding.