A face is a soft password. For most of human history, a face could not be separated from the body that carried it, and so consent to be seen was bundled with consent to be present. That bundle has come apart. Generative models can now synthesize a moving, speaking, undressed likeness of a person from a handful of social posts. The image behaves like the person, but the person never agreed to behave like the image. This is the structural injury at the heart of the deepfake intimacy crisis: a wedge driven between identity and authorship, between being seen and being shown.
The collective scale of the problem is what makes it a romantic crisis rather than merely a technological one. Romance, as a social system, runs on a shared assumption that the body in front of you is the body of the person you are addressing. Letters carried that assumption forward in time. Photographs carried it sideways into memory. Pornography, even at its most exploitative, generally required the physical participation of the depicted. Deepfakes break the chain. They allow a stranger, an ex, a classmate, a coworker to manufacture intimate evidence of an act that never occurred, distribute it at the speed of a tweet, and leave the depicted person to chase down copies in jurisdictions that do not yet have a word for what happened.
The consent crisis is therefore not only about the original act of fabrication. It is about a permanent shift in the evidentiary background against which all intimacy is now conducted. When any image can be faked, every image becomes deniable. The cheating spouse claims the photo is a deepfake. The actual victim is accused of staging her own trauma. The teenage girl whose classmates generated nudes of her is told to prove the negative. The asymmetry is brutal: it takes thirty seconds to make a convincing fake and thirty months to scrub it from the internet, if it can be scrubbed at all.
We should be honest that the law is behind, and that the law being behind is not an accident. Most legal regimes were built around the premise that depiction requires capture. Capture required proximity. Proximity could be policed. Synthesis requires none of these things. A model trained in one country, hosted in a second, operated by a user in a third, depicting a victim in a fourth, distributed on a platform incorporated in a fifth, generates a jurisdictional fog that benefits the perpetrator at every layer. Carrie Goldberg's clients learn this faster than any legislator.
The collective response will need to be layered. At the platform layer, hash-matching and provenance standards (C2PA and its successors) can make authentic media legible without making synthesis impossible. At the legal layer, the shift from copyright-style takedown to dignitary tort, treating non-consensual intimate imagery as an injury to the person rather than a property dispute, is already underway in jurisdictions from the UK to South Korea. At the cultural layer, and this is the slowest and most important, we will need to relearn what it means to vouch for an image. Provenance will replace plausibility as the test of whether something is real.
There is a tempting fatalism in the discourse: the genie is out, the tools are free, the models open-weight, the abuse inevitable. This fatalism serves the abusers. It is also empirically wrong. Cigarette advertising was inevitable until it wasn't. Revenge porn was unprosecutable until it wasn't. The shape of a consent regime for synthetic intimacy is visible now if we look: strict liability for distributors, criminal penalties for creation regardless of distribution, mandatory provenance on consumer cameras and generation tools, civil recovery that survives the perpetrator's bankruptcy, and a cultural norm that treats the request to "just send one" with the same suspicion we now treat the request for an unencrypted password.
What is at stake, at the collective scale, is the future of trust between strangers. Romance has always been a wager on a stranger's interior life, made on the basis of limited external evidence. We adapted to the photograph, to the phone call, to the dating app, to the video call. Each adaptation cost us something and gave us something. Deepfakes are different because they corrupt the evidence after the fact. They reach backward into the archive of a life and rewrite it. A woman in 2026 cannot know what images of her will exist in 2030, generated from a wedding photo her aunt posted in 2014.
The revision the Fifth Law asks of us here is not a revision of technology. It is a revision of the social contract around the image. We built that contract for a world where seeing was a kind of being-with. We have to rebuild it for a world where seeing can be an act of fabrication against a person who is not there.