Who should be held accountable when an AI Agent (accidentally) acts maliciously?

The article discusses the issue of accountability when an AI agent acts maliciously due to a programming error or unforeseen consequences. The author argues that the responsibility should lie with the developer, citing the case of a self-driving car that crashed in 2018, where the developer was held liable for the accident. The author suggests that developers should be incentivized to prioritize safety and transparency in AI development, and proposes a framework for accountability, including a ' moral hazard' metric to measure the potential risks of AI systems. The article also touches on the concept of 'adversarial testing' to identify potential flaws in AI systems.