By Peter Witten and Line Horndal Hjørne, AAU Communikation and Public Affairs
Photo: Colourbox

You’ve probably experienced receiving a text message or an email trying to trick you into clicking a risky link or opening a fake website. In an instant, your phone or computer is infected with malicious software, your passwords are stolen, and you might even be scammed out of large sums of money.

Cybercrime is rapidly advancing and becoming harder to detect. Hackers have long been using AI, and digital threats are becoming increasingly sophisticated.

Now, the research project AI:DEFENCE from Aalborg University aims to develop new methods to detect and stop advanced cyberattacks, while also making AI technology more secure.

Combining knowledge

AI:DEFENCE is led by two AAU researchers: Professor Johannes Bjerva from the Department of Computer Science and Assistant Professor Qiongxiu (‘Jane’) Li from the Department of Electronic Systems.

Johannes Bjerva works with multilingual NLP and security in large language models, while Qiongxiu (‘Jane’) Li specializes in cybersecurity, privacy protection, and machine learning.

This combines expertise in cybersecurity and NLP (Natural Language Processing).

NLP is about understanding and working with human language, such as when we interact with a chatbot, ChatGPT, Google Translate, etc.

Signs of fraud

By analyzing language - for example in an email - AI:DEFENCE will be able to detect signs of fraud that humans or regular software might miss. With AI’s help, detecting scams should become easier.

The researchers behind AI:DEFENCE will use advanced techniques such as:

• Language analysis to detect phishing emails and deepfakes (fake messages created with AI).
• Reinforcement learning, where AI learns to defend itself by simulating attacks.
• Adversarial training, where AI is "attacked" during development to make it stronger.

Encrypted AI, ensuring decisions are made securely and privately.

Malware and deepfakes

Examples of digital threats AI:DEFENCE will focus on using AI:

• Polymorphic malware - malicious software that constantly changes to avoid detection.
• Deepfake phishing - fake messages generated by AI trying to deceive you.

By analyzing language in real time, these threats can be detected before they cause harm.

Open source

The tools developed by AI:DEFENCE will be used for things like chatbots that can’t be hijacked or manipulated, and for critical infrastructure like hospitals and energy grids, where preventing data leaks is crucial.

The systems from AI:DEFENCE will be available as server software or APIs that can easily integrate with existing security systems to upgrade digital defenses.

The tools will be open source, so both small businesses and large organizations worldwide can use them.