Deepfake intimacy and the consent crisis

Before generative models, the depicted and the depicter were yoked together by physics. To photograph a person, you had to be in the room. To film them, you had to point a lens. The depicted retained, by default, a kind of authorship over their own image, because nothing could be made of them without their being made present. Deepfakes drive a wedge between identity (whose face this is) and authorship (who made this image exist). The wedge is the injury. Everything downstream, the harassment campaigns, the school scandals, the political smears, the romantic blackmail, flows from this single ontological break. Hany Farid has spent two decades arguing that we underestimate how much social trust was quietly load-bearing on the photograph's indexical link to reality. He is right. We are now living in the aftermath of that link being severed, and we have not yet built the institutions that will let romance survive it.

A convincing fake takes minutes. A takedown takes months. A reputation takes years. A psyche, sometimes, never recovers. This asymmetry is not a bug in the current enforcement regime; it is the entire problem. Every other consent crisis in modern history (workplace harassment, marital rape, child abuse imagery) was eventually addressed by making the cost of the violation higher than the cost of prevention. Deepfakes invert this economy. The marginal cost of generating one more victim approaches zero, while the marginal cost of defending one more victim remains roughly constant and very high. Until that asymmetry is corrected, by strict liability, by mandatory provenance, by criminal exposure for creators rather than just distributors, the volume of abuse will continue to grow at the rate of compute.

The first wave of deepfake intimacy abuse did not land on celebrities. It landed in middle schools. Boys discovered free nudifier apps, fed them yearbook photos, and circulated the output on Snapchat before lunch. Henry Ajder's tracking of these tools showed traffic in the tens of millions per month. The girls depicted are eleven, twelve, thirteen. They learn, before they have had a first kiss, that their image is a substance other people can mine. This rewires the entry point into romantic life. A generation is forming its first model of intimate exposure under the assumption that exposure is involuntary and permanent. The collective downstream effect on willingness to trust, to flirt, to undress in front of a partner, will not be measurable for a decade, but it will be large.

The second-order harm of ubiquitous synthesis is the deniability it grants to genuine perpetrators. If any image can be faked, then no image is dispositive. The cheating spouse waves off the photo. The abusive politician calls the recording AI. The harasser claims his victim made the screenshots herself. Danielle Citron has called this the liar's dividend, and the dividend compounds. Each new generation of model raises the floor of plausible deniability. We are heading toward a regime in which only chain-of-custody provenance, cryptographically signed at the moment of capture, will count as evidence. Romance, which has always run partly on evidence (the text message, the photo, the letter), will have to learn to run on signed evidence or no evidence at all.

The architecture of deepfake abuse is transnational by default. A perpetrator in Romania uses a model trained in California on data scraped from Brazil, hosted on a server in Singapore, distributing through a platform incorporated in Ireland, targeting a victim in Korea. Every layer of this stack benefits from the friction between jurisdictions. National laws, even good ones like the UK's Online Safety Act or Korea's revisions to its sexual violence statutes, hit the wall of mutual legal assistance treaties that move at the speed of diplomacy. The collective response will require something closer to the Budapest Convention on Cybercrime, retooled for synthetic media, with real extradition teeth. Without it, victims are left with civil remedies against defendants they cannot find.

Mary Anne Franks and Danielle Citron have spent fifteen years pushing the legal frame from property to dignity. The early non-consensual intimate imagery statutes were drafted as copyright-adjacent torts, which treated the harm as a kind of theft. This framing failed victims because it required them to assert ownership of images they often had not consented to in the first place. The newer frame treats the body's image as an extension of bodily integrity, and non-consensual depiction (real or synthetic) as a violation of the person. This is the right frame. It scales naturally from photographs to deepfakes, because it locates the injury in the depicted rather than in the file.

The technical response to ubiquitous synthesis is ubiquitous provenance. C2PA, the Content Authenticity Initiative, and related standards embed cryptographic signatures at the moment of capture, recording the device, the time, the edits, and the chain of custody. The romantic implication is that the unsigned image will, within a decade, carry roughly the epistemic weight that an unsigned email carries today: possibly real, certainly unverifiable, never dispositive. Couples will learn to ask whether the photo their teenager received was signed. Dating apps will display capture provenance alongside the profile. This sounds dystopian until you remember that the alternative is a world in which any image of you can be weaponized by anyone with a laptop.

Carrie Goldberg's law practice documents what victims actually pay: the takedown fees, the reputation management contracts, the therapy bills, the lost jobs, the broken relationships, the years of vigilance. The aggregate cost runs into the tens of thousands of dollars per victim, and falls disproportionately on women, minors, and people whose social capital is already thin. This is the victim's tax, and it is currently uncompensated. A serious collective response would include mandatory restitution funds, paid into by platforms and model providers in proportion to the abuse their tools enable, drawn down by victims without requiring them to identify or sue the perpetrator. The analogy is the vaccine injury compensation program: socialize the residual harm of a useful technology rather than pretend it does not exist.

Sherry Turkle's clinical observations of digitally mediated intimacy now include a new category: patients whose romantic lives are organized around the fear of synthetic exposure. They do not undress in front of partners. They do not allow photographs. They scan reverse-image search weekly. The therapeutic frame is still catching up. Lori Gottlieb's columns increasingly field questions from people whose relationships have been damaged not by infidelity but by the suspicion of synthetic infidelity, a partner who believes a deepfake is real, or refuses to believe a real image is real. The clinician's job, in this new territory, is partly epistemological: helping the couple agree on what counts as evidence at all.

The platforms that host, recommend, and monetize synthetic intimate imagery are not neutral conduits. Their recommendation systems optimize for engagement, and synthetic NCII engages. Section 230 in the United States, and its analogues elsewhere, were drafted for a world in which platforms did not algorithmically amplify content. That world is gone. The collective response will require a duty of care: a legal obligation on platforms to detect, remove, and prevent re-upload of synthetic intimate imagery, enforced by regulators with the power to fine into the billions. The UK and EU are moving in this direction. The US is behind, partly because its First Amendment doctrine treats synthetic speech as expressive, a doctrinal error that will need to be corrected.

At the cultural layer, slower than law and slower than tech, a new etiquette is forming. It is becoming rude, then suspect, then disqualifying, to request intimate images from a partner without an explicit, ongoing, revocable consent framework. Young couples are negotiating image rights the way previous generations negotiated exclusivity. The vocabulary is still clumsy (sext pacts, image NDAs, deletion clauses) but the underlying move is correct: treat the image as a thing that lives on after the relationship and plan for its afterlife in advance. This is the cultural correlate of the legal dignitary turn, and it will eventually be as routine as exchanging keys.

The Fifth Law is revision: the willingness to change the rules when the world has changed under them. The rules around intimate depiction were built for a world of physical capture, and they functioned, imperfectly, for a century. That world is gone. Holding the old rules in place is not conservatism; it is abandonment of the people the rules were supposed to protect. The revision required is structural: new law, new platform duties, new provenance infrastructure, new clinical frameworks, new cultural norms. It is also personal: each of us has to revise our own assumptions about what an image is, what consent covers, and what we owe the people whose faces pass through our feeds. Romance survives this revision only if we do it together, and only if we do it soon.

Ajder, Henry, Giorgio Patrini, Francesco Cavalli, and Laurence Cullen. The State of Deepfakes: Landscape, Threats, and Impact. Amsterdam: Deeptrace, 2019.

Citron, Danielle Keats. The Fight for Privacy: Protecting Dignity, Identity, and Love in the Digital Age. New York: W. W. Norton, 2022.

Citron, Danielle Keats. Hate Crimes in Cyberspace. Cambridge, MA: Harvard University Press, 2014.

Farid, Hany. Fake Photos. Cambridge, MA: MIT Press, 2019.

Franks, Mary Anne. The Cult of the Constitution. Stanford, CA: Stanford University Press, 2019.

Franks, Mary Anne. "Drafting an Effective 'Revenge Porn' Law: A Guide for Legislators." Cyber Civil Rights Initiative, August 17, 2015.

Goldberg, Carrie. Nobody's Victim: Fighting Psychos, Stalkers, Pervs, and Trolls. New York: Plume, 2019.

Gottlieb, Lori. Maybe You Should Talk to Someone. Boston: Houghton Mifflin Harcourt, 2019.

Turkle, Sherry. Alone Together: Why We Expect More from Technology and Less from Each Other. New York: Basic Books, 2011.

Turkle, Sherry. Reclaiming Conversation: The Power of Talk in a Digital Age. New York: Penguin Press, 2015.

Chesney, Robert, and Danielle Keats Citron. "Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security." California Law Review 107 (2019): 1753-1820.

Farid, Hany. "Creating, Using, Misusing, and Detecting Deep Fakes." Journal of Online Trust and Safety 1, no. 4 (September 2022): 1-33.