When you sit down with Champie Joyce Maptue Tagne, you quickly realize she doesn’t talk about cybersecurity the way most people do. There’s no hype, no dramatic warnings, no buzzwords thrown around for effect. She speaks like someone who has spent years watching how malware behaves in the real world: quietly, persistently, and always one step ahead of defenders.
TechBullion caught up with Champie to discuss her recent peer‑reviewed publication in IEEE Access, in which she examined how large language models (LLMs) are used in malware detection. The paper, “Large Language Models for Malware Detection: A Systematic Review, Taxonomy, and Open Challenges,” has been circulating among researchers who are trying to understand whether AI can genuinely help in the fight against malicious software.
Champie didn’t mince words.
“People assume AI is going to solve malware magically,” she told us. “It won’t. But it can help us fight smarter if we understand what it can and can’t do.”
She leaned back as she said it, almost as if she’d repeated that sentence a hundred times to people who wanted simple answers to complicated problems.
Champie explained that her motivation for the study came from watching how quickly malware evolves compared to how slowly traditional detection methods adapt.
“Attackers don’t wait for academic papers,” she said. “They innovate constantly. Meanwhile, defenders are stuck with tools that sometimes feel like they were built for a different era.”
With LLMs suddenly showing an ability to read code, explain behavior, and even reason about software logic, Champie felt the field needed clarity, not excitement.
“There was so much research coming out at once,” she said. “Different methods, different models, different claims. It was becoming hard to tell what was actually working and what was just experimental noise.”
Her systematic review pulled all of that scattered work together, organizing it into a taxonomy that researchers can use to understand the landscape.
One of the most interesting parts of our conversation came when Champie talked about the difference between recognizing code and understanding malicious behavior.
“LLMs are great at pattern recognition,” she said. “They can tell you what a piece of code looks like. But malware isn’t just about what code looks like; it’s about what it intends to do.”
She paused for a moment.
“That’s where things get tricky. An AI model might flag something as suspicious, but if it can’t explain why, how do you trust it? Cybersecurity isn’t guesswork.”
Her research highlights exactly that problem: LLMs can assist, but they also hallucinate, misinterpret behavior, and sometimes produce confident but incorrect explanations.
“Confidence is not competence,” she added. “And in cybersecurity, confusing the two can be dangerous.”
Champie’s paper doesn’t just summarize research; it organizes it. She created a taxonomy that breaks down how LLMs are being used:
- classification
- code reasoning
- feature generation
- behavioral analysis
This structure helps researchers see what has already been explored and where the gaps still exist.
“We needed a map,” she said. “Otherwise, everyone is running in different directions, and the field becomes impossible to navigate.”
Her work also identifies open challenges, including evaluation issues, explainability gaps, and the risk of over‑reliance on AI in security‑sensitive environments.
Champie is clear about one thing: she doesn’t believe AI will replace cybersecurity professionals.
“People want shortcuts,” she said. “But cybersecurity is not a job you automate away. It’s a job where you use tools to make better decisions.”
She sees LLMs as potential assistant tools that can speed up analysis, highlight suspicious patterns, and help researchers understand complex code faster.
“But the final judgment,” she said, “should always come from a human who understands the stakes.”
Champie’s work on LLM‑based malware detection is part of a broader research path that includes Android security, software quality, malware analysis, and AI reliability.
All of it, she said, connects back to one question:
“How do we use intelligent systems to make the digital world safer without creating a false sense of security?”
It’s a question she believes the cybersecurity community must take seriously as AI becomes more embedded in security workflows.
As our conversation wrapped up, Champie reflected on why her study matters.
“AI is moving fast. Malware is moving fast. If research doesn’t keep pace, we’ll always be reacting instead of preparing.”
Her work provides a foundation, a structured understanding of what the field knows, what it doesn’t, and where it needs to go next.
“It’s not about predicting the future,” she said. “It’s about making sure we’re ready for it.”
And with researchers like Champie leading the conversation, the future of malware detection may be more thoughtful and more secure than the hype suggests.
