Responsible AI in Academic Research: A Competency Framework for Research Training
※ This document is a translated and summarized version of the Instats policy report, Responsible AI in Academic Research: A Competency Framework for Research Training, published in May 2026.
Key Takeaways
- Every so often while writing a paper, a question comes up: “Can I have ChatGPT handle this literature review?” “This wording was polished by AI — can I just use it as is?” The problem is that there’s nowhere to get a clear answer.
- Publishers moved quickly, converging within ten weeks of ChatGPT’s launch on a single rule: AI cannot be a co-author. But the standard researchers actually face every day — how far is too far — still amounts, in most cases, to a plagiarism check. A survey of 38 leading universities across 15 countries found only six with a mature policy addressing both research integrity and AI literacy.
- Given this institutional gap, this article lays out a five-dimension competency framework that lets researchers set their own standards — sorting AI use into four modes (search, co-author, validator, tutor) and specifying what to hand off to AI and what never to.
Introduction: The Response Gap in Academia and the Need for a Framework
Since ChatGPT-3.5’s public release in late 2022, generative AI — including large language models (LLMs) — has triggered a fundamental shift in research practice. The key actors in the academic ecosystem have responded at very different speeds and in very different ways. Major academic publishers reacted fastest and most uniformly: within just ten weeks of ChatGPT-3.5’s release, they converged on essentially the same policy — AI cannot be a co-author of a paper. National research funders, by contrast, rolled out related policies more slowly and unevenly between September 2023 and April 2026, and even then mostly limited to confidentiality during proposal review.
The most important point is the institutional lag at universities — the very places that train the PhD-level researchers who will actually use these tools. A survey of 38 leading research universities across 15 countries found that most institutions remain stuck in a category error, reducing AI to a ‘plagiarism’ problem in paper writing. Only six universities had a mature AI policy covering research integrity, AI literacy, and sound research practice. Even the 2025 revision of Vitae — the UK’s authoritative researcher development framework — did not treat AI as an independent competency, which makes the limits of the existing research-training system plain.
To close this institutional gap, this report defines generative AI as an ‘agentic research tool’ — on par with a statistical package or a data-cleaning pipeline. Building on that premise, it offers an in-depth analysis of the five-dimension competency framework and maturity grid that university research leadership should adopt, along with standards for AI use at the level of individual research tasks, the tool environment needed to support responsible use, and the key literature that underpins academic integrity.