Facebook's deepfakes policy is a step in the right direction but remains severely limited, covering only certain videos created by AI while excluding satire
It's better than nothing — but just barely — Monday, Facebook announced a new policy against “deepfake” videos.
Context & Ripple Effects
Facebook's announcement lands after months of internal signaling: Mark Zuckerberg had already argued in mid-2019 that the definition of a deepfake was too broad and needed deliberate scoping, so the narrow final policy reads as intent executed rather than drift. The immediate test case critics reach for is the distorted Nancy Pelosi video — slowed to make her appear impaired — which was never AI-generated and therefore falls outside the new rule entirely.
The policy also arrives as Facebook's peers take a different route: Twitter's draft framework proposed labeling synthetic media and warning users before sharing rather than banning it outright (opened for public input in November). The OneZero follow-up the next day sharpens the critique, arguing the satire carve-out creates a loophole bad actors can drive through.
First-order effects
- Facebook moderators gain an enforceable rule only for videos made with AI techniques meeting its narrow definition, while cheap non-AI manipulations like the Pelosi edit remain permitted content on the platform.
- The satire exemption hands political operatives a ready-made defense: misleading synthetic or manipulated clips can be reclassified as parody and stay up.
Second-order effects
- Twitter's label-and-warn approach now functions as the visible alternative in the coverage, pressuring Facebook to justify why removal-only-for-AI beats disclosure-based handling as the two platforms' policies diverge.
- Enforcement depends on detection tooling Facebook does not yet have: its own Deepfake Detection Challenge later found the winning algorithm identified fakes at just 65.18% average accuracy, meaning even the narrow ban will miss most violations at scale.
Third-order effects
- If platform-by-platform scoping continues, synthetic-media governance fragments into a patchwork where the same clip is banned on one network, labeled on another, and untouched on a third — pushing regulators toward baseline definitions of manipulated media rather than leaving scope-setting to each company.
- The gap between what policies cover and what detectors can catch points toward provenance infrastructure — watermarking and origin metadata — becoming the durable fix, since post-hoc detection at 65% accuracy cannot carry enforcement weight.
The trend: Social platforms are converging on narrowly scoped synthetic-media rules built around their own definitions and detection limits, leaving low-tech manipulation and satire-shaped deception largely ungoverned.