Synthesis of Strategies for Information Integrity and Manipulation in Adversarial Environments
Abstract
Publicly Releasable Project Abstract The central thesis of this proposal is that it is essential for autonomous systems to create a strategic advantage to operate in adversarial environments by proactively gaining information in a timely manner, protecting the integrity of their information, and manipulating the information they release to their adversaries and the environment. The proposed effort will develop theory and algorithms for synthesizing strategies that will enable three increasingly proactive mechanisms for suppressing the information leakage from the system, introducing deceptive behavior into the system s operation, and engaging in continuing interactions with potential attackers for inference and diagnostic purposes. It is structured into three complementing technical thrusts. Thrust I -- Suppression of information leakage: Reduce the likelihood or delay for the attacker to recover safety- or mission-critical information from its observation of the system s execution. The resulting algorithms will incorporate several information-theoretic metrics, e.g., entropy and Fisher information, into the synthesis of strategies for the system subject to temporal logic specifications. Thrust II -- Synthesis of deceptive strategies: Synthesize strategies for the system to make sequential decisions and emit information that deliberately misleads an adversary while the system preserves its ability to accomplish its mission. Such decisions will include design-time allocation of resources, e.g., decoys with different information content and detectability and sensors with heterogenous capabilities and quality, and run-time actions, e.g., for altering the network traffic and application usage. Thrust III -- Inference for diagnostics: Thrusts I and II focus on defense in continuing operations. Thrust III focuses on diagnostics for better defense in the future. The resulting algorithms will have both run-time and post-operation elements. At run time, they will help the system counteract the attacker in order to lengthen the duration of interactions with acceptable risk-taking, e.g., by defining and synthesizing strategies for sequential decision-making that create an effect analogous to honeypots. Post operation, they will interrogate the available data and models to infer the attacker s potential intent and the attack surfaces it may utilize. These mechanisms have the potential to deflect the attention of the attacker, push it to allocate its time and information processing resources ineffectively, and consequently take risks that it would not take otherwise. The algorithms to be developed in Thrusts I-III will help run systematic what-if studies to understand the capabilities of (hypothetical) attackers against which strategic advantage can be created.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 09, 2023
- Source ID
- W911NF2310317
Entities
People
- Ufuk Topcu
Organizations
- Army Contracting Command
- United States Army
- University of Texas at Austin