AIR Framework Research Transparency: A Critical Look at AI Disclosure in Academia
- David Ruttenberg
- 20 hours ago
- 5 min read
By Dr David Ruttenberg | June 2026 | ~1,300 words · approx. 5-minute read
Why "Did You Use AI?" Is No Longer a Simple Question
Here's a question that's quietly causing chaos in universities, journals, and labs right now:
"Did you use AI in your research, and if so, how?"
A few years ago, the honest answer was usually "no" or "a little." Today, the honest answer for many researchers involves something like:
AI‑assisted literature search.
AI‑generated first drafts, reworked substantially.
AI‑based data analysis tools.
AI as an accessibility accommodation (screen readers, caption tools, writing assistants).
The problem is that nobody has been speaking the same language about any of this—until now.
Enter the AIR (AI in Research) framework, a structured disclosure matrix that maps AI use across seven research stages (discovery, implementation, analysis, writing, publication, outreach, evaluation) and five engagement bands (from "no use" to "substantial use").
It's the closest thing academia has produced to a shared vocabulary for AI transparency—and a critical analysis of it was just published in the Springer journal AI & Society this week.
What the AIR Framework Research Transparency Model Actually Does
Think of AIR as a grid. On one axis, you have every major stage of research—from early literature searches all the way through post‑publication outreach. On the other axis, you have levels of AI involvement, from "I didn't touch it" to "AI did a significant portion of this work."
The result is a map researchers can fill out to explain, specifically and consistently, where and how AI was used. Instead of "I used AI a little," you can say "AI was used at the Analysis stage, Band 2, with all outputs reviewed and substantially revised."
That level of specificity matters for:
Peer reviewers who need to evaluate what a human researcher actually contributed.
Journals trying to set consistent AI disclosure policies.
Funding bodies assessing whether methods are reproducible.
Readers who deserve to know what they're trusting.
In a field full of vague, inconsistent, and sometimes evasive disclosures, AIR is a real step forward. The critical analysis in AI & Society acknowledges this, calling it "valuable transparency infrastructure."
But Here Are the Five Problems
The same paper does not let AIR off the hook. After running an inter‑rater reliability pilot study (15 raters, 9 scenarios) that found reasonable agreement among trained users (Cohen's κ = 0.72), the author identifies five significant limitations.
1. False precision in ambiguous practices
Some AI use cases fall messily between bands. A researcher who uses an AI writing assistant to restructure an argument but not to generate text: Band 2 or Band 3? The current framework forces a choice where the honest answer is "it depends."
2. Inadequate treatment of accessibility‑related AI use
This is the one that matters most for this community. A researcher who uses AI tools because of dyslexia, ADHD, autism, or a physical disability may end up reporting "high" AI use under a framework that treats all Band 4–5 use as inherently suspect. Their accommodation gets mistaken for misconduct. More on this in a moment.
3. Stigmatization of legitimate high‑band practices
Sometimes heavy AI use is not "cheating"—it's the methodologically honest way to handle massive datasets, multilingual sources, or large‑scale content analysis. The current band labeling risks stigmatizing these valid choices.
4. Vulnerability to adversarial compliance
Like any disclosure framework, AIR can be gamed. A researcher who wants to look transparent can fill out the matrix in technically accurate but misleading ways. The framework flags this risk without fully solving it.
5. Insufficient edge‑case guidance
Real research is messy. The current AIR matrix doesn't yet have a community‑maintained library of "here's how to classify this weird scenario"—leaving many researchers making inconsistent individual judgment calls.
The Accessibility Gap: This Is the One That Keeps Me Up at Night
Let's stay with limitation number two.
The analysis proposes a specific fix: a protected "A1‑Access" sub‑band that would distinguish AI use for disability accommodation from AI use that changes the intellectual substance of the research.
Why does this matter? Because without it, a neurodivergent researcher who:
Uses an AI writing assistant to manage executive‑function challenges.
Uses AI captioning or speech‑to‑text because of a hearing or motor disability.
Uses AI summarization tools to navigate a reading disability.
…may appear, on a standard AIR matrix, to have used AI "substantially" in ways that raise integrity flags.
In a research culture that already disadvantages disabled and neurodivergent scholars through every layer of academic gatekeeping, this is not a small technical glitch. It is a structural barrier wearing a transparency framework as a disguise.
The proposed A1‑Access sub‑band says: accommodation use is categorically different. It should be reported—it is honest to report it—but it should not trigger the same scrutiny as substantive AI‑generated content.
That's not lowering the bar. That's understanding that the bar was built for one kind of researcher.
AIR Framework Research Transparency and the Bigger Picture
This paper arrives at a moment when:
Most journals have AI disclosure policies that contradict each other.
Many researchers feel trapped between transparency and self‑incrimination.
Disabled and neurodivergent scholars face the additional bind of disclosing accommodations in systems not built to understand them.
The AIR framework research transparency model doesn't solve all of that. But the critical analysis in AI & Society is exactly the kind of rigorous, constructive engagement that moves a field forward rather than just complaining from the sidelines.
Its five refinements—boundary case designations, the A1‑Access sub‑band, separation of verification from judgment, spot‑check validation studies, and community edge‑case repositories—are not radical asks. They're the work of getting a good idea closer to a useful one.
For researchers, integrity officers, journal editors, and anyone who cares about keeping disabled and neurodivergent voices in science: this paper deserves your attention this week.
References
Ruttenberg, D. (2026). The AIR framework for research transparency: A critical analysis of stage‑specific AI disclosure in the context of accessibility and research integrity. AI & Society. https://doi.org/10.1007/s00146-026-03082-x
University of Glasgow Pathfinder. (2026). AIR: AI in Research—A framework for transparent and responsible AI use mapped to the research process [Framework resource]. https://www.linkedin.com/posts/pathfinder-at-the-university-of-glasgow
Hashtags
#AIRFrameworkResearchTransparency #AIEthics #ResearchIntegrity #Neurodiversity #AccessibilityInResearch #AIDisclosure #S2MHD #NothingAboutUsWithoutUs #CognitiveLiberty
About the Author
Dr David Ruttenberg PhD, FRSA, FIoHE, AFHEA, HSRF is a neuroscientist, autism advocate, Fulbright Specialist Awardee, and Senior Research Fellow dedicated to advancing ethical artificial intelligence, neurodiversity accommodation, and transparent science communication. With a background spanning music production to cutting-edge wearable technology, Dr Ruttenberg combines science and compassion to empower individuals and communities to thrive. Inspired daily by their brilliant autistic daughter and family, Dr Ruttenberg strives to break barriers and foster a more inclusive, understanding world.
Comments