Future Program: Securing Artificial Intelligence for Battlefield Effective Robustness (SABER)

Given the latest advances in artificial intelligence (AI), the technology has reached a level of maturity, prompting a growing desire for these technologies to be integrated into Department of Defense (DoD) systems. AI may be used to give battlefield advantage by helping improve the speed, quality, and accuracy of decision-making while enabling autonomy and assistive automation. Due to the statistical nature of machine learning, prior work has focused primarily on ensuring the robustness of AI-enabled systems at inference time due to natural degradations in performance caused by data distribution shifts, for example, from a highly dynamic deployment environment. However, as early as 2014, AI has been demonstrated as manipulable given adversary control of the input [1]. Initial work has further confirmed the theoretical risks of data poisoning [2], physically constrained adversarial patches for evasion [3], and model stealing attacks [4]. These attacks are typically tested in simulated or physical environments with relatively pristine control compared to what might be expected on a battlefield. Today, there is no known reported ecosystem that operationally assesses deployed military AI-enabled systems for adversarial vulnerabilities, so the “theoretical” adversarial AI attacks have not been practically demonstrated in operational settings. Given this, the operational security risks of AI-enabled battlefield systems remain unknown. As such, SABER will build an exemplar AI red team equipped with the necessary counter-AI techniques, tools and technical competency to operationally assess AI-enabled battlefield systems, while establishing a sustainable model for an evergreen operational AI red teaming process. Our AI red team will be operationally assessing AI- enabled autonomous ground and aerial systems that could be deployed within the next 1-3 years. To assist the AI red team, the future solicitation will seek performers who can assist in surveying, evaluating, selecting, developing, and employing state-of-the-art physical (including manufacturing/materials), adversarial AI (including digital), cyber and electronic warfare techniques and tools, or other relevant vectors for enabling the operational assessment of AI-enabled system development and deployment pipelines [5]. Additionally, the forthcoming solicitation will seek a performer to serve as an integration lead, assisting in integrating the technologies into an interoperable operational AI red teaming toolkit to equip the AI red team and enable future AI red teams for the DoD and broader U.S. Government.

Citations:

1. https://arxiv.org/abs/1412.6572
2. https://arxiv.org/abs/1804.00792
3. https://arxiv.org/pdf/2104.06728
4. https://arxiv.org/abs/2411.10023
5. https://sam.gov/opp/041e8bd594c1450dbec14a6ca580fbed/view

Artificial intelligence (AI) has reached an age of maturity, and there is a growing desire for these technologies to be integrated into systems that can be used to give battlefield advantage by helping improve the speed, quality, and accuracy of decisions in the field while enabling autonomy and assistive automation. Due to the statistical nature of machine learning, prior work has focused on the robustness of AI-enabled systems at inference time to degradations in performance due to data distribution shifts from a highly dynamic battlefield environment. However, as early as 2014, AI has been demonstrated as manipulable given adversary control of the input [1]. Initial work has further confirmed the risks of data poisoning [2], physically constrained adversarial patches for evasion [3], and model stealing attacks [4]. However, these attacks are typically tested in simulated environments, or physical environments with relatively pristine control compared to what might be expected on a battlefield. Because adversarial AI attacks have not been practically demonstrated in operational settings, the operational security risks of AI-enabled battlefield systems remain unknown. Given the above, SABER will build an exemplar AI red team that can continuously integrate and employ emerging counter-AI techniques and tools, establishing a sustainable model for an evergreen operational AI red teaming process. Our AI red team will be operationally assessing AI-enabled autonomous ground and air systems that could be deployed in the next 1-3 years. To assist the AI red team, this solicitation seeks performers who can assist in surveying, assessing, selecting, developing, and employing state-of-the-art physical (including manufacturing/materials), adversarial AI (including digital), cyber and electronic warfare techniques and tools, or other relevant vectors for enabling the operational assessment of AI-enabled system development and deployment pipelines. Additionally, this solicitation seeks performers to serve as an integration lead, assisting in integrating the technologies into an interoperable operational AI red teaming toolkit to equip the AI red team and to enable future AI red teams for the DoD and broader USG.

Citations:
1- https://arxiv.org/abs/1412.6572
2- https://arxiv.org/abs/1804.00792
3- https://arxiv.org/pdf/2104.06728
4- https://arxiv.org/abs/2411.10023