AI to AI Autonomous Cyber Warfare: Ethical and Anticipated Ethical Issues

Abstract

The phrase “AI‑to‑AI autonomous cyber warfare” refers to an emerging and/or anticipated type of digital conflict in which artificial intelligence systems engage in cyberwarfare without any direct intervention by humans. In this type of conflict AI systems will carry out the collection of data through sensing, decision‑making, and responses to adversaries as cyber conflicts unfold in cyberspace. AI to AI Autonomous Cyber Warfare needs to be examined at several levels as part of defense theory, cybersecurity research, and in anticipation of future risks related to cyber warfare. AI to AI Autonomous Cyber Warfare must be thought of as potentially related to the automation of both cyber offense and cyber defense, where both friendly and adversarial machines will interact with machines at speeds humans cannot match. In cyberwarfare humans usually set goals, define rules, and identify constraints, but AI systems in AI-to-AI warfare will execute all actions independently of human agency. The idea behind AI-to-AI Autonomous Cyber Warfare is that they are like autonomous systems employed as defense agents today—they are just more advanced and will have more authority. Currently AI systems exist on opposing sides, and they are already identifying, analyzing, and/or responding to each other. Cyber operations happen at machine speed, which is much faster than humans. Current systems of AIs may adapt strategies based on the opponent’s behavior. This has led to an environment where two or more autonomous AI agents can currently learn, escalate, and counter each other dynamically. What already exists are: (1) Automated intrusion detection systems using ML. (2) Autonomous patching or quarantine systems. (3) Malware that is self-adapting.  What is expected to exist in the future are (1) Fully autonomous offensive systems. (2) AI agents that will be able to strategically outmaneuver each other. (3) Self‑modifying agents that will be able to escalate activities without human oversight. It is currently agreed that fully autonomous AI to AI cyber warfare is a major global risk, but that it is not a current capability. The current set of risks related to the development of AI‑to‑AI autonomous cyber warfare include: (1) Speed-of-Conflict Escalation. (2) Unpredictability. (3) Accidental conflicts (4) Arms Race Dynamics. (5) Loss of Human Control. This analysis will focus on ethical and anticipated ethical issues with the development of AI‑to‑AI autonomous cyber warfare.