The Perils of Agency: How Developers Perceive, Prioritize, and Address Risks in Agentic AI Products

A study examining 35 industry developers of agentic AI products reveals a consistent pattern in risk perception and prioritization. Developers primarily link risks to the core agentic characteristics of products, such as autonomy and tool use. Their prioritization heavily favors product and business risks over broader societal concerns like employment impacts or end-user privacy, which consequently influences mitigation efforts. The study indicates a lack of maturity in addressing these agentic risks among developers.

Why it matters

The findings underscore a significant gap in how AI product developers perceive and address the broader implications of their creations, particularly concerning societal impact. This prioritization framework has direct consequences for the responsible development and deployment of advanced AI systems, necessitating strategic adjustments in risk management and ethical considerations across the industry.

Key insights

• Developers associate risks in agentic AI products with the products' autonomous nature, tool-use capabilities, and operation in real-world environments.
• Product and business risks are prioritized by developers above societal risks.
• Prioritization of product/business risks impacts developers' ability and motivation to mitigate broader agentic risks.
• Societal risks, such as job displacement and end-user privacy, receive lower prioritization.
• Developers lack maturity in addressing the identified agentic risks.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2606.15485