Don’t bring an AI detector to a deepfake fight: proving reality through multimodal provenance

Per LessWrong, we're losing the battle against deepfakes—not because fakes are too good, but because we're fighting the wrong war. The real problem isn't viral propaganda; it's something subtler: as fakes become indistinguishable from reality, even authentic evidence becomes deniable. A defendant could simply say 'deepfakes exist, so this video could be fake'—introducing reasonable doubt regardless of whether it was actually manipulated. Rather than trying to detect generated content, LessWrong proposes inverting the approach: default-distrust all images and videos, but require proof of authenticity through cryptographic signing during capture. Combined with platform accountability and legal consequences for untagged generated content, this could restore trust in real evidence without playing an arms race we'll inevitably lose.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton