This research examines whether Large Language Models can effectively fact-check news content they generate, revealing important limitations in AI self-verification capabilities.

Key Findings:

• LLMs perform better at fact-checking national and international news than local stories
• Models handle static information more accurately than dynamic or rapidly changing facts
• There are significant gaps in LLMs' ability to detect their own generated misinformation
• These limitations pose critical security concerns for news verification in AI-generated content

Security Implications: As AI-generated news becomes increasingly common, understanding these verification limitations is essential for developing robust safeguards against misinformation and maintaining information integrity in digital ecosystems.

Fact-checking AI-generated news reports: Can LLMs catch their own lies?