Canada’s three federal research funding agencies — the Natural Sciences and Engineering Research Council (NSERC), the Canadian Institutes of Health Research (CIHR), and the Social Sciences and Humanities Research Council (SSHRC) — collectively distribute billions of dollars annually to researchers at universities and institutions across the country. The tri-council system is the backbone of publicly funded academic science in Canada. It is also built on a foundational assumption: that the applications it receives represent the genuine scholarly thinking, methodological judgment, and intellectual labour of the researchers who submit them.
That assumption is now under serious strain. As AI agents — autonomous systems capable of conducting literature reviews, synthesizing research directions, drafting narratives, and iterating on outputs with minimal human intervention — become increasingly accessible, the question is no longer whether they will be used in grant writing. They already are. The question is what happens to the integrity of peer review and public funding allocation when the tools doing the writing are indistinguishable from the researchers themselves.
What AI Agents Actually Do — and Why It Matters for Grants
AI agents differ from simple generative AI tools like a chatbot used to polish prose. They can be instructed to pursue a goal across multiple steps: search funding agency databases for priority areas, review recently funded projects to identify gaps, synthesize an applicant’s publication record into a coherent research narrative, and produce a polished draft that conforms to agency formatting requirements. Some can iterate based on reviewer feedback from prior cycles. The output is not a rough draft — it is a strategically calibrated document.
This creates several compounding problems for grant systems. First, it divorces the application from the applicant’s genuine thinking at the time of writing. Second, it potentially advantages researchers who have access to sophisticated agent infrastructure — or the technical sophistication to deploy it — over those who do not, introducing a new axis of inequity into an already competitive system. Third, and most consequentially, it challenges the premise of peer review itself: if reviewers are evaluating an agent’s synthesis rather than a researcher’s vision, the signal they are assessing has fundamentally changed.
Canada’s Tri-Council Is Not Ready
NSERC, CIHR, and SSHRC have each issued guidance on AI use in research contexts, but that guidance has largely focused on disclosure in publications and data integrity — not on the grant application process itself. The agencies have not, as of the most recent funding cycles, established binding standards for AI use in the preparation of funding proposals. Voluntary disclosure norms, where they exist, are inconsistently applied and largely unenforceable.
The absence of clear rules is not a neutral position. It effectively permits unrestricted AI agent use in grant preparation while placing compliant researchers — those who choose to disclose or limit their AI use — at a potential competitive disadvantage relative to those who do not. This is a governance failure that compounds over every funding cycle it goes unaddressed.
Canada’s situation is not unique globally, but it carries particular weight here. The tri-council system funds research that informs federal policy, trains the next generation of Canadian scientists, and positions Canada in international science diplomacy. The legitimacy of that system depends on public and institutional trust that funded researchers earned their awards on the merits of their ideas.
The Detection Problem Is Real — and Probably Unsolvable Alone
A tempting policy reflex is to call for AI detection tools as the solution. That reflex should be resisted. Detection technology for AI-generated text is unreliable, produces false positives at rates that would create serious due-process problems in high-stakes funding decisions, and is already being outpaced by more sophisticated generation methods. Researchers writing in English as a second language are disproportionately flagged by existing tools, adding a discriminatory dimension to an already blunt instrument.
Detection cannot be the primary line of defence. It may have a supplementary role, but policy frameworks that rely on it as a cornerstone will fail — and will fail inequitably.
What Federal Science Policy Must Do Before the Next Cycle
The tri-council agencies, ideally in coordination with the federal Office of the Chief Science Advisor and Innovation, Science and Economic Development Canada, need to move on several fronts before the next major funding cycle.
- Establish mandatory, standardized disclosure requirements for AI tool use in grant preparation — not as a punitive measure, but as a condition of submission that enables agencies to understand the scope of the phenomenon and build evidence for further policy.
- Develop a working definition of permissible and impermissible AI assistance in grant applications, distinguishing between using AI to check grammar, using it to generate literature syntheses, and using autonomous agents to produce substantive proposal content.
- Revisit the application process itself — including interview components, staged submissions, or post-award verification mechanisms — to reintroduce direct evidence of researcher thinking that agents cannot fully substitute.
- Invest in equity analysis: who has access to advanced agent tools, and how might unrestricted use entrench existing disparities between well-resourced and under-resourced institutions?
- Engage the research community transparently, including the universities and their research offices, many of which are already grappling with these questions internally without federal guidance to lean on.
The Deeper Stakes
Canada has made significant public commitments to AI leadership — through the Pan-Canadian AI Strategy, through investment in institutes like Mila, the Vector Institute, and Amii, and through the positioning of Canadian AI research as a national economic and strategic asset. There is an uncomfortable irony in AI undermining the credibility of the very funding mechanisms that sustain that research ecosystem.
The threat posed by AI agents to grant integrity is not hypothetical or distant. It is occurring now, in the current funding environment, without adequate policy infrastructure to manage it. NSERC, CIHR, and SSHRC have an opportunity to lead on this — to establish frameworks that other national funding bodies will look to as models. The alternative is allowing the problem to metastasize until a high-profile case of agent-assisted fraud forces a reactive and likely disproportionate response.
Federal science policy should not wait for that moment. The next funding cycle is not far away.
Related InsightTrack Analysis
- AI Agent Orchestration Frameworks for Workflow Automation
- Agentic AI Benefits and Risks for Canadian Enterprises
- Local AI Deployment in Canada: Business Benefits
Source
New AI agents pose ‘existential threat’ to how grants are awarded
