Skip to main content

Emerging Cyber Technologies Research

Research Focus Areas

Graphic of LLM representation

Large Language Models

ECTR both uses and defends large language models (LLMs) in the security domain. We are working to bring the benefits of LLMs to the security professional by operationalizing these models for more sophisticated tasks in threat intelligence gathering, malware analysis, and reverse engineering. Yet, LLMs also present attackers with a new attack surface, so we are exploring ways to harden LLMs against data poisoning, prompt injection, data extraction, and other adversarial exploitations.
Graphic representing AI model robustness

AI Model Robustness

We develop platforms to ensure and enhance AI model robustness, which is especially critical when these tools are used in national security applications. Robust AI plays a vital role in safeguarding national infrastructure, from communication networks to energy grids, against cyber espionage and sabotage.
Graphic representing reinforcement learning for robust cyber defense

Reinforcement Learning for Robust Cyber Defense

Our experts are exploring ways to use reinforcement learning to help defenders identify, contain, and recover from attacks. This approach, a type of machine learning where algorithms learn to make decisions by interacting with an environment, enables AI systems to learn from real-time cyberattacks and adapt their defenses dynamically. Such systems can simulate numerous attack scenarios, continuously learning and improving their defense mechanisms.
AI generated image representing AI security

Explainable AI

The ECTR group deploys the latest developments in explainable AI (XAI) to aid cyber defenders. XAI aims to make the often-opaque decision-making processes of AI systems more transparent and understandable to humans. This is crucial in cybersecurity, where understanding the 'why' behind a decision or a prediction is as important as the decision itself. XAI fosters trust among users and stakeholders, ensuring that AI-driven security measures are not just effective, but also accountable.