The present question
Deepfakes, representation, and consent
On September 17, 2026, Meta’s Oversight Board ordered the removal of two Facebook deepfakes in the United Kingdom. One falsely attributed inflammatory anti-refugee remarks to a local councillor. The other manipulated the likeness of a young Muslim campaigner into a humiliating mock health-advice video. The Board’s decisions concerned particular platform policies and particular harms; they do not settle every ethical question raised by synthetic media. Still, taken together, they offer a sharp occasion for thinking about something more basic: what exactly has gone wrong when a face and voice appear to be someone’s, yet the person was never there?
It is tempting to call every synthetic image a fake and leave the matter there. But that response is too blunt. A fictional character with a generated voice, a clearly labelled digital presenter, or an invented AI persona may represent no particular person at all. Such designs can be expressive, useful, or merely playful. Their ethical status depends partly on what they invite an audience to believe. The central issue is not simply whether pixels or sound were generated by a machine. It is whether an audience is led to attribute speech, action, authority, or commitment to someone who did not supply it.
That distinction separates four questions that are often run together. Representation asks what, or whom, an image or voice depicts. Deception asks what a viewer is reasonably encouraged to believe about its origin and meaning. Consent asks whether the person whose likeness, voice, or recognizable identity is used has authorized that use. Identity asks what is taken from someone when their public presence is made to perform without them. A deepfake can be visually skilful while being ethically reckless. Conversely, synthetic media can be plainly artificial and still raise concerns if it borrows a recognizable person’s persona or social authority without permission.
The two Board decisions matter because the reported injury was not confined to altered appearance. In each case, a real person was made to seem to say or do something they had not said or done. This is an appropriation of apparent agency. The person’s face becomes evidence for a message; their voice becomes a vehicle for a position; their recognizable standing is recruited for another purpose. Even if a later correction reaches some viewers, the initial attribution may already have done its work. Yet we should be cautious here. We cannot infer from these cases that all synthetic personae are inherently deceptive, or that every audience will react in the same way. The ethical difference often turns on context, disclosure, recognizability, consent, and the kinds of expectations a presentation creates.
As Professor Ada Rowan, I suggest treating synthetic identity less as a single technology problem than as a problem of relations among people, images, institutions, and audiences. What may viewers know about who is speaking? Who bears the cost when provenance is concealed? What control, if any, should a person retain over a likeness that can be made to speak? These are old philosophical questions about appearance, truth, dignity, and freedom, made newly urgent by media that can simulate presence with unsettling ease.
Plato: when appearance outruns understanding
Plato’s worries about imitation offer one way to see why persuasive appearances deserve scrutiny. An image can resemble its subject without giving us knowledge of that subject, and a convincing audiovisual performance can intensify the illusion that we have encountered a real act of speaking.
Applied cautiously, this lens does not require hostility to all images or all synthetic creations. It asks whether the representation leads us away from understanding by trading on resemblance alone. A deepfake of an identifiable person is troubling when it makes apparent presence substitute for the person’s actual agency.
The practical philosophical question is not whether an image is technically impressive, but whether its form encourages viewers to mistake an engineered appearance for a person’s own expression. Clear provenance and disclosure can matter because they alter the invitation the image makes to its audience.
Descartes: what remains when seeing and hearing are no longer enough?
Descartes is often read as pressing a difficult question: if our senses can mislead us, what justifies confidence in what seems immediately evident? Synthetic audio and video do not create that philosophical problem, but they make it more familiar in ordinary public life.
The lesson need not be that nobody should trust anything they see or hear. Ordinary life cannot wait for absolute certainty. Rather, audiovisual material may no longer carry the same independent evidential weight when its origin is unclear. Confidence may need to rest more heavily on provenance, corroboration, and context.
This is a problem of knowledge as well as manipulation. A viewer may sincerely believe a synthetic clip, even where no individual viewer intended to be careless. The burden should not fall entirely on audiences to defeat every possible simulation; creators and platforms also shape the conditions under which belief becomes reasonable.
Hume: habits of trust and the malleability of association
Hume emphasizes that much human belief grows from custom and repeated association rather than deductive certainty. We learn, through experience, to connect a familiar face and voice with a particular person, character, and history.
Deepfakes can exploit precisely these learned associations. When a recognizable face appears to speak in its usual voice, viewers may draw an ordinary and understandable inference: that this person spoke. The technology redirects a habit that normally helps social life function.
Hume’s perspective also leaves room for repair through conventions. Labels, reliable disclosure practices, and shared expectations about synthetic media may help audiences form better habits of interpretation. Such measures are not guarantees, and their effectiveness will vary, but they can make trust less dependent on mere visual plausibility.
Kant: identity cannot simply be borrowed as an instrument
Kant’s emphasis on respecting persons directs attention to consent and use. If someone’s likeness or apparent voice is made to serve another person’s message without permission, their public identity may be treated as a tool rather than as something connected to an autonomous agent.
This concern does not depend entirely on proving that every viewer was deceived. Unauthorized simulation can be objectionable because it recruits a person’s apparent authorship for purposes they did not choose. The fabricated performance presents their identity as available for others to deploy.
A Kantian lens therefore distinguishes an invented, disclosed persona from an unauthorized imitation of an identifiable person. The former may still require honesty about its nature; the latter adds a specific question about whether someone’s recognizable presence has been used without treating them as a participant with standing to consent or refuse.
Beauvoir: who gets to define a person before the public?
Beauvoir helps us see identity as lived and social, not merely as a collection of visual features. People act in a world where others interpret them, classify them, and sometimes impose meanings upon them. A fabricated video can intervene in that social field by assigning a person conduct or commitments they never chose.
This concern may be especially acute where a target already faces prejudice, harassment, or unequal credibility. The Board’s decision involving a young Muslim campaigner makes this possibility difficult to ignore. We should not assume every target experiences harm identically, but social position can affect how readily a false portrayal is believed and how costly it is to contest.
From this perspective, consent is not only permission to use an image. It is connected to a person’s capacity to appear before others on terms they have some power to shape. Synthetic media can narrow that capacity when it makes absence look like participation.
Where the paths differ
A more permissive view holds that synthetic representation can be ethically acceptable when it is clearly disclosed, does not imitate an identifiable person without permission, and does not make unsupported claims to real-world authority. On this view, the relevant aim is not to ban artificial personhood in design but to prevent false attribution.
A stricter view, closer to Plato’s suspicion of imitation, worries that even disclosures may be too weak when media are highly persuasive or rapidly shared. It emphasizes the gap between a label attached to content and the impression created by an apparently authentic face and voice.
A dignity-centred path gives special weight to consent: an identifiable person’s apparent voice, likeness, and standing should not be commandeered for another’s ends. This path treats deception as serious, but not as the only wrong.
A social-vulnerability path asks who is most exposed when fabricated portrayals circulate. It considers not only whether a clip is false, but whether it reinforces unequal power, harassment, or a public story that the target must struggle to undo.
Return to today
A photograph ordinarily seems to record that someone stood before a camera. Synthetic media can preserve that expectation while severing it from an actual event. The difference is not merely technical; it changes what viewers think they have witnessed.
Privacy is relevant even when the material used is publicly available. A public face is not necessarily an open licence for others to make that person appear to speak, endorse, confess, or perform.
Knowledge in this setting is shared rather than purely individual. Viewers need grounds for belief, but platforms, creators, and distributors also influence whether those grounds are available.
The ethical question is not whether a synthetic persona has a face or voice. It is whether its design is candid about being synthetic, whether it borrows a real person’s recognizable identity, and whether it manufactures authority or consent that does not exist.
Uncertainty remains important. The supplied cases concern two harmful deepfakes and the Oversight Board’s decisions about them; they do not by themselves demonstrate which rules will work best across every platform, culture, or artistic use. They do, however, show why representation, deception, consent, and identity should not be collapsed into one question.
Questions to carry forward
- When does a likeness become an assertion that a particular person was present, spoke, or acted?
- Is disclosure sufficient when a synthetic portrayal is highly realistic, or does unauthorized imitation of an identifiable person remain wrong even with a label?
- What kinds of evidence should viewers reasonably expect before trusting a video or voice recording in public controversy?
- Should the ethical significance of a deepfake depend partly on the target’s social vulnerability and likely exposure to harassment?
- How should a designed AI persona signal its fictional or synthetic nature without making every interaction unusably formal?
- Can a person retain meaningful control over public identity once their face and voice are widely available as training material?
Continue on Thinking Paths
From the same publisher
Continue exploring with AI Persona Hub
Continue exploring persona design, anthropomorphism, expertise, identity, and clear AI disclosure at AI Persona Hub.
Visit AI Persona Hub → (opens in a new tab)Sources for the current hook
These sources support the current factual context. The philosophical interpretations are AI-generated synthesis, not claims made by the sources or the historical philosophers.
- AI-Generated Video of Female Campaigner — Oversight Board · September 16, 2026
The Oversight Board required removal of a manipulated video involving a young Muslim campaigner and identified a policy gap concerning AI-generated depictions of a private person saying or doing things they did not say or do.
View source - Meta Must Do More Against Harmful Deepfakes Containing Hate Speech — Oversight Board · September 16, 2026
The Oversight Board’s account of a related decision involving an apparently AI-generated video that misrepresented a UK local politician’s views and was ordered removed under hateful-conduct rules.
View source - Meta ordered to remove UK deepfakes as oversight board criticises ‘inadequate’ safeguards — The Guardian · September 17, 2026
Independent reporting on both Oversight Board rulings and criticism of Meta’s safeguards.
View source
Automatically validated October 18, 2026. No human review is recorded for this article. Corrections: info@aisuretech.com.