Detecting Machine-Generated News Using Fine-Tuned Transformers
Open Access DepositedPhishing is a leading source of cybercrime, and attacks have become more sophisticated where social media is growing as an effective attack vector. The emergence of generative artificial intelligence (GenAI) has raised questions about its application. Democratizing the technology has made using GenAI easier to use. Malicious actors are finding creative ways to use GenAI to generate fake news, impersonate individuals, spread disinformation, and create complex social engineering campaigns for phishing attempts. This research focused on fine-tuning large language models (LLMs) for classifying machine-generated fake and human-created real news articles. The dataset for this research was a collection of pre-generated GPT-2 news articles, along with real news articles gathered from the internet using Common Crawl. This data combined with the latest popular LLMs such as Llama, Gemma, and Mistral constructed new detection models for fake news. LLMs Meta Llama 2 70b, Meta Llama 3 70B, Mistral AI Mixtral 8x 7B, Mixtral AI Mixtral 8x 22B produced the best results achieving an Accuracy of greater than 99.9%. This research further explained the challenges and outcomes of using commercial-off-the-shelf (COTS) hardware to successfully create and evaluate multiple fine-tuned transformer models. This work will be useful for the cybersecurity information technology industry in its continuing efforts to finding solutions for detecting malicious content generated by machine.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.