National Security AI Threat Framework - A hybrid / human centred model for AI-enabled adversarial operations

Hey everyone!

Very happy to share my newest framework for the assessment of AI-enabled adversarial operations.

In today's rapidly evolving technological landscape, the vast majority of the workforce doesn't "speak AI," nor is versed on critical aspects of modern intelligence or other contemporary adversarial operations. From analysts to strategy architects, researchers and  policymakers, bridging these worlds is also a costly and complex endeavor. In their own way, the intelligence community (IC) also is required to see technologies in dual-use capacities, facing the unique challenge of maintaining in-house narratives for both defensive and offensive capabilities, often executed with defensive intent, but offensive nonetheless. 

It is in this context that the **National Security AI Threat Framework** (NSAITF) bridges these worlds together for comprehensive analysis.

Designed to meet both digital (or cyber) and non-digital activities and operations, and recognizing the increasingly prevalent use of AI in automating several aspects of such operations, the NSAITF provides a high-level model both offensive and defensive AAI/AML (Adversarial Artificial Intelligence/Adversarial Machine Learning) scenarios.  It synthesizes technical, strategic, operational, and ethical considerations into a cohesive model, enabling a multidimensional assessment that extends beyond technical specifics. This framework provides a structured perspective on AI-enabled operations, emphasizing human-centric analysis and illuminating the interdisciplinary impacts on the social fabric.

By transcending siloed analysis, the NSAITF empowers users to navigate the complexities of AI-enabled operations across various domains, including cyber, political, economic, and cognitive spaces. 

Its strength lies in qualifying the strategic breadth and operational weight of these operations, offering perspectives to both IC members and other stakeholders. The NSAITF integrates PMESII (Political, Military, Economic, Social, Information, Infrastructure) and other frameworks concepts into one, combining capabilities of threat intelligence frameworks, impact assessment models, risk evaluation tools, consequence analysis frameworks, and recovery planning modelsinto a hybrid assessment framework. This framework is highly capable of assessing both objectives and outcomes, allowing for forward planning or the reverse engineering of operations, greatly enhancing its utility in the real world.

A key feature of the NSAITF is its emphasis on the human perspective behind adversarial operations. It aims to empower a more informed society, where students, journalists, researchers, and the IC can understand and address the societal effects of these activities, including those domestic in nature, and their side effects, both short and long term. The framework calls for a realistic and foundational approach towards fairness and proportionality in these operations, fostering a informed society.

The framework have its own GitHub repository with its full taxonomy and instructions for use here: https://github.com/antoniomaxai/NSAITF/

It has also been tested against 23 out of 29 Cyber Law collection scenarios from NATO's Cooperative Cyber Defence Centre of Excellence (CCDCOE). This approach enabled both qualitative assessment and quantitative analysis of the societal impacts of cyberwarfare operations. The results of this evaluation are available in a public spreadsheet at: https://docs.google.com/spreadsheets/d/1FeekXROr04drN8gcQYSVbv28M3wKvIqJlTgzGLz0bmQ/edit?usp=sharing

The NSAITF sees intelligence and adversarial operations as a reality of our times. It is well capable of covering cyber treats, misinformation, election interference, among many other types of threats, with a normalized taxonomy that can always illustrate human impacts of these operations.

It is also very easy and intuitive to use, enabling both qualitative and quantitative analysis out-of-the-box. 

I invite you to check its repository, use it with your own use case, and please: let all improve it together, this is just its first version, and I hope we can improve it over time, have alternative versions, it is an open model, you may use and customize it as you see fit!

The image on this post is its full table diagram, where you can see all the NSAITF covers and its groups. The shared Google Sheets document above showcases just one way to perform its analysis! Would love to read your feedback!

 

 

NSAITF Table Diagram
The full NSAITF table diagram. The framework enables a hybrid threat assessment for AI-enabled operations.
Sildid
Adversarial AI Adversarial Machine Learning National Security