Releases: mzyphur/responsible-ai-in-research-training
Release list
v1.2.4 — Canonical AI-assistance disclosure + Word-compatibility rebuild
What changed
- AI-assistance disclosure. Appendix D now carries the canonical Instats Policy Series disclosure wording: it names the AI tool families used in preparing the report and states that the author is responsible for its content.
- Rebuild for Word compatibility. Rebuilt with additional document-validity checks layered on top of the v1.2.1 bookmark fix and the v1.2.3 chart colour conversion.
No substantive content changes from v1.2.3. Evidence, framework, recommendations, charts, and DOI are unchanged.
Downloads
| Format | Asset |
|---|---|
| Microsoft Word | report.docx |
report.pdf |
|
| HTML | report.html |
Citation
Zyphur, M. J. (2026). Responsible AI in Academic Research: A Capability Framework for Research Training. Instats Policy Series, v1.2.4. https://github.com/mzyphur/responsible-ai-in-research-training. ORCID: 0000-0003-3237-7892.
Full version history: CHANGELOG.md.
v1.2.3 — Chart PNG RGBA→RGB (Mac Word compatibility)
Fixes the v1.2.2 DOCX not opening in Mac Microsoft Word. The Figure 1 chart PNG is now RGB (no alpha channel), so the image embedded in the DOCX is byte-equivalent to a clean RGB image.
No visual change to the chart and no prose changes from v1.2.2.
v1.2.2 — Figure 1 chart redesign
Figure 1 redesigned: right-side callout panel lists the six Class D universities by date (no more overlapping leader-line callouts); inline lane labels with publishers/funders/universities timing positioned cleanly; real date X-axis (Nov 2022 through May 2026) replaces abstract month numbers. Publication-quality, no overlapping text.
No prose changes from v1.2.1. Same evidence base, framework, four-way standard, societal-stakes paragraph, and DOI (10.61700/t31oy23grr).
DOCX bookmark regression check: all 53 bookmarks remain Word-compatible after rebuild. The v1.2.1 Mac Word fix is sticky.
See CHANGELOG.md for full v1.2.0 + v1.2.1 + v1.2.2 entries.
v1.2.1 — Fix Mac Word docx incompatibility
The DOCX now opens in Mac Microsoft Word. The root cause was a Word-incompatible bookmark name produced when the colon in the Figure 1 chart anchor ({#fig:timeline}) was URL-encoded to fig%3Atimeline. Mac Word's stricter document validator requires bookmark names to contain only letters, numbers, and underscores; LibreOffice tolerates %, which is why the defect was not caught in local round-trip testing.
Fix: {#fig:timeline} → {#fig-timeline} (hyphen instead of colon). Rebuilt; all 53 bookmark names are now Word-compatible.
No content changes from v1.2.0. Same title, framing, body, evidence, framework, chart, and DOI (10.61700/t31oy23grr).
v1.2.0 — Define 'responsible' + societal stakes
Adds two paragraphs to the report: an explicit definition of "responsible" in §1.2 (the four-way standard — ethical, valid, reproducible, transparent — with a defining clause for each, and the necessary-together claim that no three substitute for the fourth) and a societal-stakes paragraph in the Conclusion (universities as epistemic infrastructure; the public's interest in knowledge it can trust; the four-way standard as the precondition for that trust).
The Conclusion now reads as problem → stakes → solution.
No empirical content changed; same evidence base, framework, chart, and DOI (10.61700/t31oy23grr) as v1.1.x. See CHANGELOG.md for the full entry.
v1.1.2 — Cover-abstract closing sentence
Small follow-on to v1.1.1. Adds one closing sentence to the cover-page abstract:
This report describes the significant opportunities and problems that agentic generative AI creates for research, setting out a capability framework for research training in the 21st century.
The sentence plants "agentic generative AI" at the cover (where it was previously only introduced inside the executive summary), previews the deliverable (the capability framework), echoes the subtitle, frames the report as opportunities-AND-problems (not anti-AI), and anchors the temporal stakes (21st century). The README headline paragraph gains the same sentence.
No substantive content changed from v1.1.1: same title, framing, body, evidence base (38 universities; 14 funders + ERC; 18 publishers; 13 frameworks; 11 AI tool classes), Figure 1 chart, and DOI 10.61700/t31oy23grr.
See CHANGELOG.md for the v1.1.0 / v1.1.1 / v1.1.2 entries.
v1.1.1 — Comprehensive framing redraft
Comprehensive framing redraft from v1.0.0. Same evidence base, same DOI (10.61700/t31oy23grr); new title ("Responsible AI in Academic Research"), SPSS-analogy reframe of LLMs as agentic research tools, new Figure 1 (regulatory-response-timeline chart), trimmed front matter, all section closes reframed. See CHANGELOG.md for the full v1.1.0 + v1.1.1 entries.
v1.0.0 — Responsible AI in Research and Research Training
Responsible AI in Research and Research Training: A Global Capability Framework for University Research Leadership
Instats Policy Series · INSTATS-PS-2026-04 · 2026-05-18
DOI: 10.61700/t31oy23grr (locked into CITATION.cff)
Cite as: Zyphur, M. J. (2026). Responsible AI in Research and Research Training: A Global Capability Framework for University Research Leadership. Instats Policy Series, v1.0.0. DOI: 10.61700/t31oy23grr.
Headline finding
Of thirty-eight top-tier doctoral universities surveyed across fifteen countries and jurisdictions, only six (about one in six) have AI policies that extend past plagiarism into research integrity, AI literacy, and examiner discipline. The world's major academic publishers issued a substantively identical no-AI-co-author policy across the sector within ten weeks of ChatGPT-3.5's public release. The United Kingdom's canonical PhD-researcher-development framework, refreshed in 2025, did not surface AI as a competency at all. The regulatory cascade has hardened at the publishing tip and the funding head; it has stopped almost completely at the institutional layer that actually trains researchers.
What the report ships
- Five-dimension capability framework (Part 2): human-in-the-loop discipline; responsible use in practice; tooling that promotes responsible use; AI-literate humans; institutional benchmarking grid.
- International evidence base (Part 3): 14 national research funders across 11 countries plus the EU's European Research Council; 38 top-tier doctoral universities across 15 countries and jurisdictions; 13 existing capability frameworks; 18 publishers plus 3 preprint servers.
- Maturity grid (Part 4 + Appendix A): a five-by-four self-diagnostic tool combining the five dimensions with four maturity levels (absent / nascent / established / leading).
- Audience-tiered recommendations (Part 5): concrete actions for five tiers — executive research leadership; faculty deans of research; deans of graduate schools; higher-degrees committees; peak bodies for graduate students.
- Labour-vs-judgement task taxonomy (Appendix G): a working catalogue of nineteen PhD-level research tasks classified as labour (AI may augment with verification) or judgement (AI may not substitute).
Review
The manuscript went through independent adversarial fact-checking, a reader-persona panel across the five audience tiers, multi-pass copy-editing, and a final voice review before release. Findings from those reviews were resolved across v0.5.1, v0.5.2, v0.6.0, and v1.0.0.
Source repository
The canonical manuscript is at drafts/report.md. Primary-source evidence dossiers supplied the underlying evidence base; the report's footnotes carry the source trail.
Build artefacts (HTML / DOCX / PDF) will follow in a v1.0.1 patch.
License
Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).