Security for intelligence.
Intelligence for security.

Research scope

The growing autonomy and interactivity of generative AI systems are reshaping their security and privacy threat landscape, particularly as large language models are embedded in agentic systems with persistent state, external tools, and the ability to act in dynamic environments.

We identify and mitigate emerging risks—from uncovering hidden threats through black-box adversarial attacks, to systematically characterizing security properties, and ultimately enabling continuous protection at runtime.

Research scope

The rapid advancement of generative AI has increased the realism, accessibility, and scale of synthetic media, creating growing concerns around content authenticity, harmful content, and the safety of generative models.

We study both post-generation risks and model-level safeguards: detecting and proactively defending against DeepFakes, understanding attempts to evade detection and safety mechanisms, and helping generative models recognize and suppress harmful concepts.

Research scope

The reasoning, coding, and tool-use capabilities of large language models and autonomous agents are opening new opportunities for vulnerability research in complex software, firmware, and networked systems that underpin critical infrastructure.

We combine program understanding, vulnerability reasoning, evidence-driven validation, and autonomous attack execution to improve the efficiency, depth, and autonomy of security analysis.