Are LLMs Vulnerable to Preference-Undermining Attacks (PUA)? A Factorial Analysis Methodology for Diagnosing the Trade-off between Preference Alignment and Real-World Validity

Documenti analoghi

• Exploring Advanced Methodologies in Security Evaluation for LLMs
  di: Huang, Jun, et al.
  Pubblicazione: (2024)
• Injecting Falsehoods: Adversarial Man-in-the-Middle Attacks Undermining Factual Recall in LLMs
  di: Fastowski, Alina, et al.
  Pubblicazione: (2025)
• PATCHEVAL: A New Benchmark for Evaluating LLMs on Patching Real-World Vulnerabilities
  di: Wei, Zichao, et al.
  Pubblicazione: (2025)
• QUIC-Exfil: Exploiting QUIC's Server Preferred Address Feature to Perform Data Exfiltration Attacks
  di: Grübl, Thomas, et al.
  Pubblicazione: (2025)
• JailPO: A Novel Black-box Jailbreak Framework via Preference Optimization against Aligned LLMs
  di: Li, Hongyi, et al.
  Pubblicazione: (2024)