Likely AI-generated. Pangram flags 89% 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
17 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
AI safety evaluation has made significant progress. However, these frameworks remain predominantly Western, English-language, and tech-centric — creating socio-technical blind spots in what gets measured and what impacts get examined.
This grant funds 6 months of full-time research to complete two papers addressing both sides of this gap:
What are this project's goals? How will you achieve them?
Specific actions:
Complete and submit a sole-authored peer-reviewed position paper developed from Beyond Technical Compliance (BTC) IV ("Whose Safety? Linguistic Hegemony as Designed Fragility"), targeting journals such as AI & Society or Ethics and Information Technology. The paper will move beyond documenting structural gaps to articulating concrete adoption pathways for Beyond Technical Compliance (CAAF).
Complete a co-authored paper with Prof. Akira Tokuhiro (Ontario Tech University) on AI-driven structural capability loss (he leads, draft in progress), with the aim of presenting at a conference in Canada or Japan..
How will I know if I've achieved it:
One manuscript submitted to a peer-reviewed journal, one co-authored paper completed and presented, within 6 months.
Path to impact:
AI governance standards are crystallizing in 2026-2027 (EU AI Act implementation, AISI evaluation protocols, frontier model safety cases). These frameworks can be amended later, but dimensions built in from the start shape what gets resourced, measured, and prioritized. Peer-reviewed publications in this window reach decision-makers while standards are still being written. These papers establish the analytical foundation for a subsequent phase: developing CAAF into a working audit protocol, potentially with Prof. Tokuhiro as co-applicant.
Relation to fund goals:
This is AI governance research addressing a structural gap that current evaluation frameworks were not designed to cover, in behavioral accountability across cultural contexts, linguistic and cultural integrity, what evaluation metrics themselves measure, and whether affected communities have genuine input into system design. It is early-stage, individual-led, and the type of work TAIF describes as its target: research that helps the transition to transformative AI go well by reducing risks that existing institutions are not yet equipped to see.
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.