This research explores using multi-modal large language models as deepfake detectors, benchmarking the latest AI systems against traditional detection methods.

• Evaluates 12 cutting-edge multi-modal LLMs including GPT-4o, Gemini Flash 2, Claude 3.5/3.7
• Tests capabilities across increasingly sophisticated synthetic media
• Compares performance against specialized deepfake detection algorithms
• Provides insights into the strengths and limitations of using general-purpose AI for security applications

As deepfakes become more realistic and prevalent, this research offers critical insights for cybersecurity professionals seeking to deploy AI-based detection methods to combat misinformation and digital forgery.

Can Multi-modal (reasoning) LLMs work as deepfake detectors?