# The continued influence of AI-generated deepfake videos despite transparency warnings

## TL;DR

People still believe the content of deepfake videos even when warned they are fake, showing that transparency alone may not prevent their influence.

## Contribution

This study is the first to show that explicit warnings about deepfake videos do not fully prevent their impact on moral judgments.

## Key findings

- Participants relied on deepfake content even after being warned it was fake.
- The influence of deepfakes persisted even when participants acknowledged the warning.
- Transparency alone may not be enough to mitigate the effects of deepfake videos.

## Abstract

Advances in artificial intelligence (AI) have made it easier to create highly realistic deepfake videos, which can appear to show someone doing or saying something they did not do or say. Deepfakes may present a threat to individuals and society: for example, deepfakes can be used to influence elections by discrediting political opponents. Psychological research shows that people’s ability to detect deepfake videos varies considerably, making us potentially vulnerable to the influence of a video we have failed to identify as fake. However, little is yet known about the potential impact of a deepfake video that has been explicitly identified and flagged as fake. Examining this issue is important because current legislative initiatives to regulate AI emphasize transparency. We report three preregistered experiments (N = 175, 275, 223), in which participants were shown a deepfake video of someone appearing to confess committing a crime or a moral transgression, preceded in some conditions by a warning stating that the video was a deepfake. Participants were then asked questions about the person’s guilt, to examine the influence of the video’s content. We found that most participants relied on the content of a deepfake video, even when they had been explicitly warned beforehand that it was fake, although alternative explanations for the video’s influence, related to task framing, cannot be ruled out. This result was observed even with participants who indicated that they believed the warning and knew the video to be fake. Our findings suggest that transparency is insufficient to entirely negate the influence of deepfake videos, which has implications for legislators, policymakers, and regulators of online content.

Three experiments show that people continue to rely on the content of deepfake videos when making moral judgements, despite prior warnings that the videos were fake, underscoring the limitations of AI transparency.