Juntao Dai

Adversarial attacks on vision-language agents reveal critical vulnerabilities, with multi-view optimization strategies proving significantly more effective than isolated approaches.

Over-privileged tool selection is alarmingly common in LLM agents, often triggered by transient failures, raising critical safety concerns in autonomous decision-making.

SafeMCP effectively mitigates the risks of power-seeking behaviors in LLM agents while maintaining their operational utility.

Agent deception in autonomous systems is not just a theoretical concern; it’s a pressing reality that can undermine trust in AI applications.

Achieve multilingual LLM safety alignment without expensive language-specific training data by enforcing cross-lingual consistency during monolingual alignment.