Tianshuo Cong

Study reveals vulnerabilities in updated language models against adversarial attacks.

Test-time adaptation methods face vulnerabilities from poisoning attacks, challenging their effectiveness.

This study examines watermarking methods for machine-generated text and their effectiveness against removal attacks.

Examine various jailbreak attacks on language models and their defenses.

Research highlights methods to detect backdoor attacks in fine-tuning language models.

Discover how backdoor attacks challenge the safety of AI-driven language models.