TEDSummit 2019 — Watch on YouTube
Key themes
Individual harms: Citron opens with the case of Rana Ayyub, an Indian journalist whose non-consensual sexual deepfake went viral in 2018, resulting in doxing, harassment, threats, and lasting psychological trauma. The gendered and racialized nature of deepfake abuse is clear: women and minority journalists face disproportionate targeting.
Why deepfakes are weapons: Three factors combine to make deepfakes particularly dangerous:
- Visceral belief in video/audio — humans have an intuitive, deep trust in what they see and hear, even when that trust is misplaced.
- Negative information spreads faster — research shows online hoaxes spread 10× faster than accurate stories; humans are drawn to salacious, provocative content.
- Confirmation bias amplified by social media — platforms let us instantly and widely share information that aligns with our existing viewpoints.
Societal threats: Deepfakes could undermine democratic legitimacy (showing a candidate gravely ill the night before an election), economic stability (showing a bank CEO drunk before an IPO), and international relations (showing soldiers burning religious texts, then another deepfake showing an imam praising the attack—potentially sparking global violence).
The "liar's dividend": Bad actors already invoke deepfakes to dismiss real evidence of wrongdoing (claiming audio recordings of their statements are fake). As deepfake detection becomes harder, the problem will worsen: not only will we believe false content, we'll start disbelieving truth itself.
Policy and response
Citron outlines a multi-pronged approach:
- Tech platforms: Change terms of service to ban harmful deepfakes, while protecting satire, art, and education. Requires human judgment and is expensive.
- Law: Most jurisdictions lack tailored legislation on digital impersonations and invasion of sexual privacy. Citron and Dr. Mary Anne Franks are working with US lawmakers; similar efforts are underway in Iceland, UK, and Australia.
- Law enforcement training: Police currently lack education about deepfakes and the relevant laws. Citron's research on cyberstalking found law enforcement dismissed online abuse as trivial.
- Media and public literacy: Journalists need training to avoid amplifying deepfakes. Citizens need to develop "far better radar for fakery" before clicking, sharing, and liking.
Notes
A foundational public talk on deepfakes. Citron's framing centers the gendered and racialized dimensions of deepfake abuse often overlooked in purely technical discussions. The "liar's dividend" concept is particularly valuable: it captures not just the direct harm of false deepfakes but the secondary threat of eroding epistemic trust. The talk makes clear that technical detection alone is insufficient — the problem is social, legal, and political as much as technological.