By There’s a video of a public official saying something they didn’t say. It was made with publicly available software, by someone without specialized skills, in a few hours. It looks real enough to fool a quick glance. It gets shared twenty thousand times before anyone flags it.
That scenario is not hypothetical. It’s been happening for a while. And the technology that makes it possible is getting easier to use and harder to detect at roughly the same pace.
What Deepfakes Can Actually Do Now
Two years ago, spotting a deepfake video was possible with careful attention — unnatural blinking, edge artifacts around the face, audio that didn’t quite sync. The tells were there if you looked. They’re less reliably there now. The quality of synthetic video and audio has improved faster than detection methods have. There are documented cases of synthetic audio being used in financial fraud — voice-cloned executives authorizing wire transfers. Synthetic images have appeared in news contexts presenting as real photographs of events that didn’t happen.
The capability to create convincing synthetic media of real people saying and doing things they never said or did is now broadly accessible — not just to nation-state actors or well-resourced operations, but to individuals with a laptop and an internet connection.
Where the Real Danger Lies
The threat isn’t uniformly distributed, and it’s worth being precise about where the genuine risk concentrates.
Election periods are high-risk. The window between when synthetic media goes viral and when it can be credibly debunked is often long enough to influence news cycles, social media conversations, and potentially voter perception. A well-timed deepfake video of a candidate, released close enough to an election that fact-checking can’t catch up, could have real effects.
Financial fraud is already documented. Voice cloning has been used to impersonate executives in business email compromise attacks that have resulted in significant financial losses. This isn’t speculative — it’s a current threat that fraud investigators are dealing with.
Non-consensual synthetic intimate imagery — deepfake pornography using real people’s likenesses — is a serious harm affecting primarily women, including teenagers. Some states have passed laws specifically criminalizing it. Federal legislation has been proposed but not passed. The enforcement challenges are significant because the platforms where this content appears and the actors who create it often operate across jurisdictions.
The Broader Epistemic Problem
Beyond specific incidents, deepfakes contribute to a more diffuse problem: they make it rational to be more uncertain about whether any video or audio is real. Once the possibility of convincing synthetic media is established, even genuine content can be dismissed as fake. This creates what researchers call the “liar’s dividend” — the ability to deny real evidence by asserting that it might be synthetic.
We’re not fully in that world yet. But the trajectory is toward one where establishing the authenticity of video or audio requires verification infrastructure that doesn’t yet exist at scale — forensic chain of custody, cryptographic signing, provenance tracking. Building that infrastructure is technically possible. Whether it gets built before the problem outpaces our ability to trust media is not guaranteed.
What Helps
Media literacy — the habit of checking sources, looking for corroborating coverage, being appropriately skeptical of shareable content that provokes strong emotional reactions — helps at the individual level. Most viral misinformation, including deepfakes, relies on people sharing without verifying. Slower, more deliberate engagement with emotionally charged content breaks that chain.
Platform-level interventions have been inconsistent. Some platforms have developed policies requiring labeling of AI-generated content. Enforcement of those policies has been uneven. Detection tools that platforms could use to flag potentially synthetic content exist but are imperfect and can be circumvented.
The honest situation is that the technology is moving faster than the social and regulatory response. The gap is real and it’s being exploited.