Accountable Agents in Software Engineering: An Analysis of Terms of Service and a Research Roadmap

A new vision paper by Christoph Treude, presented at the 3rd ACM International Conference on AI-powered Software (AIware 2026), examines the accountability frameworks underlying today's AI coding assistants and autonomous agents. By systematically comparing the Terms of Service (ToS) from major providers, Treude identifies a clear trend: tool vendors consistently offload responsibility for code correctness, safety, and legal compliance onto the developers using their products. The analysis also reveals notable divergences in how providers handle indemnification, data reuse, and acceptable use policies, creating an uneven landscape where developers cannot easily predict their liability.

Treude argues that these policy documents were designed for earlier, less autonomous tools and are poorly aligned with the current reality of agent-mediated workflows where AI can independently generate, modify, and recommend code. To address this, the paper outlines a research roadmap focused on modeling responsibility, designing governance artifacts, developing tooling that supports accountability, and conducting empirical studies of developers' perceptions and practices. The work underscores an urgent need for the software engineering community to establish clearer norms and regulations around AI-generated code, especially as autonomous agents become more prevalent in production environments.