This research introduces SINCon, a defense mechanism that protects rumor detection systems from being compromised by malicious content injected using large language models.

• Addresses a critical vulnerability where LLMs can be used to generate fake messages that manipulate rumor detection systems
• Proposes a mechanism that identifies and neutralizes the influence of artificially injected content in message propagation trees
• Demonstrates significant improvement in rumor detection robustness against sophisticated AI-powered attacks
• Offers a practical solution for maintaining the integrity of social media analysis tools as LLMs become more accessible

As AI tools become more widespread, securing content analysis systems from manipulation becomes essential for reliable information integrity on social media platforms.

SINCon: Mitigate LLM-Generated Malicious Message Injection Attack for Rumor Detection