New audio and video technologies that can easily recreate any person's likeness will make fake stories harder to detect
What we saw in the 2016 election is nothing compared to what we need to prepare for in 2020. — Less than a month after Donald Trump was improbably elected … Tweets: @emilybell See also Mediagazer Tweets: Emily Bell / @emilybell : Today I have been emailing this to everyone. Good luck, fact checkers http://twitter.com/... See also Mediagazer
Context & Ripple Effects
Written less than three months after the 2016 election, this Vanity Fair piece — which Emily Bell was emailing around newsrooms — is the early-warning shot in what became a seven-year arc of coverage: it argues that software able to recreate any person's likeness will outpace fact-checking before the 2020 cycle. At the time, the concern rested on emerging face-swap and voice-cloning tools, with no name yet attached to the technique.
The follow-on coverage validates the alarm and names it: by late 2018 Wired was mapping how AI-generated fake audio and video could be aimed at elections, and by 2019 CNN reported DARPA funding detection research while flagging a subtler danger — convincing fakes give cover for dismissing real footage as fake. By 2023, Fast Company sources were warning that pairing deepfake audio and video with large language models like GPT would turbocharge scammers and propagandists, and the Financial Times entered 2024 with experts bracing for a calamitous deepfake moment in high-stakes elections.
First-order effects
- Fact-checkers and newsroom verification desks — the audience Bell was forwarding this to — inherit a new burden: every piece of audio or video showing a public figure now requires provenance checks before publication, because the tools to fabricate a plausible likeness are becoming trivially easy to use.
- Platforms that distributed fake news at scale in 2016, Facebook chief among them per later research comparing 2016 and 2018, become the natural amplification layer for synthetic media, since their reach is precisely what makes a fabricated clip valuable.
Second-order effects
- A detection arms race emerges on the government side: DARPA's fake-detection research, reported in 2019, signals that national security agencies treat synthetic media as a defense problem, pulling funding toward forensics rather than prevention.
- The 'liar's dividend' kicks in — once audiences know fakes exist, politicians and officials gain a ready-made excuse to dismiss authentic recordings as fabricated, degrading the evidentiary value of all video regardless of whether any fake appears.
Third-order effects
- If the pattern holds from these data points, synthetic media stops being a standalone threat and becomes a component: the 2023 warnings about combining deepfake audio, virtual avatars, and LLMs point toward automated, personalized disinformation pipelines that no single fact-checking workflow can keep pace with.
- Election integrity infrastructure shifts from post-hoc debunking toward provenance and authentication standards — the direction the 2019 detection research and the recurring pre-election alarms heading into 2020 and 2024 both imply, though whether detection can ever match generation speed remains genuinely unresolved.
The trend: Synthetic media is evolving from a novelty editing trick into a standing election-integrity threat, with each cycle's coverage raising the stakes faster than detection capability closes the gap.