The Integrity Paradox: An AI Expert Used AI to Write About AI—Without Telling Anyone

Cath Ellis holds an important job at Western Sydney University: she oversees quality and integrity across the institution. So when she recently wrote an opinion piece for The Sydney Morning Herald arguing that universities should embrace artificial intelligence, readers had no idea that Ellis had actually used AI to write the article itself.

According to The Sydney Morning Herald, Ellis fed 40,000 words of her own research materials into Microsoft's Copilot—a large language model, or AI system trained on vast amounts of text—and had it generate the piece. Neither Ellis nor her university told the newspaper that AI had played this role. This creates an awkward tension: the person hired to make sure everyone follows the rules used a tool she was publicly promoting, without being transparent about it.

How the University Responded

Western Sydney University defended Ellis's decision, describing her approach as "edge thinking and innovative" that prepares students for a world where AI is everywhere. The university acknowledged that AI generated the article and that it was based on Ellis's earlier work.

But this institutional support raises a practical problem. Academic journals already require scholars to disclose conflicts of interest, funding sources, and how they conducted their research. When academics write for mainstream newspapers, though, those standards are much looser. Nobody has clear rules yet about whether—or how—to tell readers when AI was involved.

Who Is Cath Ellis?

Ellis started her current role as Pro Vice-Chancellor for Quality and Integrity in November 2024. Before that, she held senior positions at multiple universities across Australia and the UK, including the University of Wollongong, University of Sydney, and UNSW. In 2019, Times Higher Education named her one of their People of the Year specifically for her work preventing academic fraud and exposing contract cheating—a practice where students pay others to write their assignments for them.

This background makes the current situation particularly striking. Ellis built her reputation on maintaining academic standards and preventing unauthorized help with academic work. Now she's at the center of a case about disclosure.

What Makes This Different From Old Debates

Universities have grappled before with questions of credit and honesty when research assistants, collaborators, or professional writers contribute to work published under a professor's name. But the current situation has a new element: the scale and speed of AI.

When Ellis uploaded 40,000 words into Copilot, she was giving the AI a massive amount of her own work. The AI then reorganized and synthesized it into a new piece. This is different from using AI to write something from scratch based on a simple question. It blurs an old line between what counts as "original work" and what counts as "revised work." If you feed your own decades of research into an AI system that rearranges and rewrites it, is that your writing? Is it someone else's? The answer isn't obvious.

Why This Moment Matters

The broader context here is that we're in an early stage of figuring out how to handle AI in scholarly work. Ellis's case is essentially a test run for how universities will handle situations where faculty members use these tools in writing that's meant for public consumption.

For Western Sydney University to say "this is fine, this shows we're innovative" sends a signal to other universities about what's acceptable. It suggests the institution is comfortable with AI integration even when transparency standards matter most. That choice will likely influence how other universities write their own AI policies for their staff.

There's an additional layer of irony here. The field of academic integrity—Ellis's own professional world—now faces questions about how to stay credible while embracing technology that scrambles what we mean by authorship. An expert in integrity rules becoming a case study in disclosure ethics is the kind of contradiction that tends to accelerate broader institutional reckoning.

What Happens Next

The most likely outcome is that this case will push universities and publishers to develop clearer rules about AI disclosure. Right now, we have a patchwork: academic journals increasingly require authors to say if they used AI, but newspaper opinion sections don't have those standards yet. That gap will probably close.

The technical detail matters too. Ellis didn't just ask an AI a question and publish the answer. She used a specific approach—feeding her own corpus of work into the system rather than relying on the AI's general training. Future conversations about AI use will probably need to distinguish between different types of assistance: light editing, significant rewriting, or synthesis of existing work are not all the same thing. A simple yes-or-no question about whether "AI was used" may not capture enough.

As AI becomes more routine in academic work, cases like Ellis's will test whether the ethical frameworks we have can adapt to what these tools actually do. The answer will likely shape how fast—and how openly—universities move toward AI integration in scholarship meant for public audiences.