SEI Names Seven Cyber Research Priorities as AI Reshapes Attacks and Defense

AI changes the threat picture

In a blog post dated October 20, 2026, SEI authors Bill Scherlis and Greg Touhill describe a shift over the past year. According to them, cyber adversaries have started to use AI to carry out a wide range of attacks, helping them accelerate, optimize and scale up their techniques. They write that agentic AI lowers the work factor for savvy adversaries and lowers the barrier to entry for new entrants.

The authors say the extent, tempo and sophistication of attacks are already increasing, and that diverse sectors are being targeted. They point to the Salt Typhoon attacks on multiple prominent telecommunications providers and, more recently, attacks on water systems attributed to Iran.

The same technology also helps defenders, according to the post. Large language models (LLMs) are now highly capable at rapidly finding vulnerabilities and developing candidate repairs. In operational settings, AI can support more effective and rapid triage in security operations centers.

The framework and its seven areas

The report was led by William Scherlis, special advisor to the director of the SEI, and Greg Touhill, director of the SEI's CERT Division. It presents 37 highest priority opportunities for advancement that, in the authors' words, hold great potential cyber benefit for national security. The seven areas are:

• Analytic and Operational Tradecraft for Cybersecurity
• Securing AI-Based Systems and Workflows
• Cyber-Physical System Security
• Cybersecurity for Complex Integrated Systems
• Insider Threat and Human, Systems Interaction
• Secure Engineering and Mission Confidence
• Modeling and Simulation in Support of Security

The team says it did not aim to be fully comprehensive. Instead it selected topics against explicit criteria, looking for the greatest potential impact on cyber operations and envisioned cyber capabilities in the Department of War, the intelligence community, the Department of Homeland Security and other federal agencies. The topics focus on technical challenges that are significant for national security and civil infrastructure, where progress is likely to make the greatest difference, and where CERT and the SEI can build collaborative relationships and realize both immediate and longer-term benefits.

Tradecraft: keeping pace with adversaries

The first area covers operationally focused work that combines computational and human elements to keep systems running securely under active threat. The authors say tactics, techniques and procedures must constantly evolve alongside adversary methods and advances such as AI-based system elements, highly distributed designs and modern heterogeneous hardware.

Issues flagged as pressing include event analysis for networked and distributed systems, data retention, analysis of adversary tactics, techniques and procedures, reverse engineering of malware, and secure operation of very large systems. On data retention, the post notes that instrumentation usually generates a large amount of highly granular data and organizations often cannot retain all of it. Research opportunities listed include efficient and confident analysis and response, red-team operations, finding and diagnosing weaknesses, rapid creation and validation of software, and cyber hygiene for AI-based systems.

Securing AI systems themselves

The second area addresses the security of AI. The authors note that modern AI rests on neural networks with billions, or even trillions, of parameters. Analysis of those parameters has shown partial success at smaller scale but has not been sufficient to support a principled understanding of outcomes at LLM scale. In their words, it is generally not possible to make confident judgments that very large models behave in particular ways.

For that reason, they say, a universal cybersecurity rubric for AI-based systems is presently infeasible. They suggest that security goals should focus on intended use cases, guiding designers on when and how to use AI and how to evaluate whether a system is fit for use. For agentic AI, they argue scrutiny should fall on the harnesses and plug-in systems that connect an LLM with external tools and networks, because these expose new attack surfaces and need principled engineering and immutable logging.

What it means

The framework is aimed at leaders in cyber research, operations and national security missions, who the authors say must focus effort where it can quickly bolster the country's cyber safety. The full report is available online with references to the technical literature. The excerpt available covers only the first two areas in detail, so specifics on the remaining five are left to the report itself.