Deepfakes & Synthetic Media
The use of AI-generated synthetic media for disinformation, fraud, and propaganda, and the challenges of detecting this technology.
OVERVIEW
Deepfakes are AI-generated synthetic media that can convincingly depict people saying things they never did. The technology has been used for non-consensual pornography, political disinformation, financial fraud, and celebrity impersonation. Detection technology improves but lags behind generation. The 'liar's dividend' allows people to claim real evidence is a deepfake. Regulation is nascent.
KNOWN FACTS
Deepfake videos of politicians have circulated online
Financial fraud using deepfake audio has been documented
Non-consensual deepfake pornography is the most common use
Detection accuracy drops for novel generation methods
The liar's dividend has been documented in court cases
CLAIMS
Deepfakes pose an existential threat to democratic discourse and evidence
The technology has been used in political disinformation campaigns
Deepfake detection technology cannot keep pace with generation
Regulation is insufficient to address the harms
The liar's dividend allows people to claim real evidence is a deepfake
EVIDENCE FOR
Deepfake videos of politicians have circulated online
Financial fraud using deepfake audio has been documented
Non-consensual deepfake pornography is the most common use
Detection accuracy drops for novel generation methods
The liar's dividend has been documented in court cases
EVIDENCE AGAINST
Most deepfake disinformation is detected before going viral
Detection technology is improving rapidly
The threat may be overstated; most disinformation is text-based
Synthetic media has legitimate creative applications
Media literacy is the best defense against deepfakes
OPEN QUESTIONS
No open questions recorded.
SOURCES
TIMELINE
First convincing deepfake videos appear
Deepfake used in political interference attempt
AI-generated audio deepfake in $25M fraud