Credibility attacks do not enhance the impact of deepfake warnings
4 affiliations
- 1School of Psychological Science, University of Western, Perth, Australia
- 2School of Psychological Science, University of Bristol, Bristol, England
- 3Department of Psychology, University of Potsdam, Potsdam, Germany
- 4Public Policy Institute, University of Western, Perth, Australia
Abstract
Synthetic deepfake videos are a realistic form of misinformation, which appear to show someone saying or doing something they never did. The societal risks posed by such fakeries have motivated legislation mandating the labelling of AI-generated content, but initial evidence suggests that even explicit warnings are insufficient to eliminate deepfake influences on viewer perceptions. Across two experiments (total N = 2,500), we examined the impact of a deepfake video in which a witness makes a criminal accusation, and we tested whether undermining the credibility of the witness could enhance the effect of a specific warning flagging the video as a deepfake. Two methods for reducing witness credibility were tested: presenting the witness with a foreign (vs. native) accent and explicitly introducing them as an untrustworthy (vs. trustworthy) character. Deepfake exposure significantly increased guilt judgements concerning the person being accused in the video; a deepfake warning only partially reduced this impact. When an explicit witness discreditation was added to the warning, effectiveness remained unchanged. Findings demonstrate the pervasive impact of known deepfakes, highlighting the need to combine or develop new interventions to combat their influence.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-64243-y.
Related
- Defending against deepfake speech attacks via feature dropout and dropin regularization in large audio language models.
- Assessing the impact of a standardised appearance and health warnings on e-cigarette appeal
- Investigating the Impact of Low-Rate Error-Correction Codes on Device-Cloning and Reverse-Engineering Attacks on Physical Unclonable Functions
- Impact of Ransomware Attacks on Small and Medium Enterprises (SMEs)
- Impact of seed node position on network robustness under localized attacks
BibTeX
@article{spearing2026credibility,
title = {{Credibility attacks do not enhance the impact of deepfake warnings}},
author = {Emily R. Spearing and Kaitlin Norgard and Simon Clark and Stephan Lewandowsky and Ullrich K. H. Ecker},
journal = {Scientific Reports},
year = {2026},
month = aug,
doi = {10.1038/s41598-026-64243-y},
url = {https://doi.org/10.1038/s41598-026-64243-y},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Credibility attacks do not enhance the impact of deepfake warnings
AU - Emily R. Spearing
AU - Kaitlin Norgard
AU - Simon Clark
AU - Stephan Lewandowsky
AU - Ullrich K. H. Ecker
PY - 2026
DA - 2026/08/05
JO - Scientific Reports
DO - 10.1038/s41598-026-64243-y
UR - https://doi.org/10.1038/s41598-026-64243-y
AB - Synthetic deepfake videos are a realistic form of misinformation, which appear to show someone saying or doing something they never did. The societal risks posed by such fakeries have motivated legislation mandating the labelling of AI-generated content, but initial evidence suggests that even explicit warnings are insufficient to eliminate deepfake influences on viewer perceptions. Across two experiments (total N = 2,500), we examined the impact of a deepfake video in which a witness makes a criminal accusation, and we tested whether undermining the credibility of the witness could enhance the effect of a specific warning flagging the video as a deepfake. Two methods for reducing witness credibility were tested: presenting the witness with a foreign (vs. native) accent and explicitly introducing them as an untrustworthy (vs. trustworthy) character. Deepfake exposure significantly increased guilt judgements concerning the person being accused in the video; a deepfake warning only partially reduced this impact. When an explicit witness discreditation was added to the warning, effectiveness remained unchanged. Findings demonstrate the pervasive impact of known deepfakes, highlighting the need to combine or develop new interventions to combat their influence. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-64243-y.
SN - 2045-2322
ER -