What People See (and Miss) About Generative AI Risks: Perceptions of Failures, Risks, and Who Should Address Them

The public has significant gaps in understanding how AI systems actually fail: The survey found that people are often unaware of specific failure modes—recurring patterns of breakdown in how AI systems are built and used—that lead to real-world harms, suggesting that general concerns about AI risks don't translate into understanding concrete problems.

Real-world incidents are more effective than abstract warnings for building AI literacy: The researchers grounded their assessment in scenarios based on publicly reported incidents rather than hypothetical risks, finding this approach more effective at helping people understand and reason about GenAI failures in contexts they actually encounter.

People may have misaligned views about who should be responsible for addressing AI risks: The survey assessed not just what risks people perceive, but also their beliefs about which stakeholders (companies, government, users, etc.) should address them—suggesting potential disagreements about accountability that could complicate governance efforts.

AI literacy interventions should match how people naturally encounter and think about AI: Rather than relying on technical explanations, the authors argue that effective AI safety education should be designed around the everyday contexts where people actually use generative AI, making it more relevant and actionable.

A validated measurement tool now exists for tracking public AI risk awareness over time: The survey instrument was validated with subject matter experts and could be used to monitor how public understanding of GenAI failures evolves, providing a baseline for assessing the effectiveness of literacy and governance initiatives.