Deepfake

Deepfake covers a spectrum: swapping a face into existing footage, cloning a voice from a short sample, or generating video of a person from scratch. What used to require a studio now takes consumer tools and a small amount of source material — a voice can be cloned from seconds of audio scraped from anywhere the person has spoken publicly.

The honest framing is that detection is not a solved counter-measure and is unlikely to become one; detectors and generators improve against each other. The more durable response is provenance — cryptographically signing content at capture and carrying that signature through editing, which is the approach behind content credentials. That establishes what is authentic rather than trying to catch everything that isn't.

The most common real-world harm is not political disinformation but fraud: a cloned voice authorising a payment, or a fabricated video call impersonating an executive. Organizations that rely on hearing a familiar voice as authentication have an urgent, concrete problem, and it is addressable with process rather than technology.

The secondary harm is subtler — as fabrication becomes plausible, genuine evidence can be dismissed as fake. That erosion of the default trust in recordings affects everyone, not only targets.

The practical defense is procedural: verify consequential requests through a separate channel, and never treat a voice or a face as proof of identity. For published media, look for content credentials rather than relying on your own eye.