Much of the public concern around deepfakes has focused on abuse, particularly non-consensual intimate imagery, political misinformation, and the erosion of trust in audio-visual evidence. These issues are real and can have serious implications.
…
But beyond personal and political harm, deepfakes introduce a broader epistemic challenge: they destabilize the credibility of recorded evidence. When people can no longer be sure what’s real, trust in journalism, democratic institutions, and even scientific data begins to erode.
One of the most pressing challenges is the potential misuse of generative AI to produce scientific data, medical imagery, or entire datasets that appear authentic but are fabricated.
A 2025 PNAS article warns that researchers, companies, or regulators may use generative models to fabricate results that appear methodologically sound. These synthetic datasets might then be presented as the outcome of actual experiments in an attempt to mislead the scientific community.
Related on the SLP
The risks include:
- Irreproducible results
- False confidence in findings
- Privacy breaches
…
This raises urgent questions about professional trust, the evidentiary standards in science, and the long-term credibility of academic publishing.
