# Proof of Human Human sovereignty in a network of machines Volume 29 · Edition 1 · 2026-08-17 Category: A · Sovereignty (https://www.thierry-gilgen-ict.ch/library/sovereignty) Canonical: https://www.thierry-gilgen-ict.ch/field-notes/proof-of-human On 30 June 2026, the Swiss Federal Office of Justice postponed the introduction of the electronic identity. The original date had been December 2026. No replacement date was announced. The reasons were unusually specific. Developments in artificial intelligence had created additional challenges for the remote process used to issue the credential. Faces can be synthesised. Video can be manipulated. Devices can be compromised. The office named two risks in particular: malware on the end-user device during online issuance, and the detection of deepfakes. Security, it said, would take precedence over the original timetable. The underlying trust infrastructure is still expected to become operational in the first half of 2027. The close reading writes itself. Machines have become too good at pretending to be humans — at least good enough that a national identity programme paused at the point where a person has to be recognised as a person, before any credential exists. The Swiss e-ID was supposed to provide a trustworthy way to prove who somebody is online. That still matters. The delay did not primarily concern cryptographic signatures, hardware keys or revocation. It concerned the first step: establishing that the applicant is a real person, present, and not a generated replica. The problem is larger than one Swiss government project. For most of the history of the internet, digital identity has tried to answer a question: **who are you?** The machine age introduces an earlier one. **Are you human?** --- ## The assumption we never designed The internet was built for people. Not exclusively — bots, crawlers and automated systems have existed almost from the beginning. Underneath most of its institutions, though, sat an implicit assumption: the meaningful actors were human. A person wrote the message, created the account, submitted the application, expressed the opinion, clicked the button, entered the conversation. We built security systems around identifying those people. Passwords authenticated accounts. Identity documents authenticated names. CAPTCHAs attempted to distinguish humans from scripts. The distinction was usually obvious enough that we did not need an architecture for it. That assumption is disappearing. A machine can now write the message, conduct the conversation, generate the photograph, imitate the voice, operate the browser and complete increasingly complex sequences of actions. More importantly, it can do so at scale. A human can create one convincing identity. A machine can potentially operate thousands. This changes the nature of digital trust. The problem is no longer merely impersonation. It is **synthetic participation**. --- ## Identity begins before identity Consider the issuance of a digital identity. Once issued correctly, the credential can be extraordinarily secure. Cryptographic signatures can establish that it is genuine. Secure hardware can protect keys. Revocation mechanisms can invalidate compromised credentials. All of this happens after a more fundamental decision has already been made. Someone had to decide whom to issue the credential to. If a machine successfully impersonates a person during that first process, strong cryptography does not repair the mistake. It preserves it. The credential can be perfectly authentic while the original claim was false. This is why identity systems distinguish between authentication and **identity proofing**. Authentication asks: is this the same entity that received the credential? Identity proofing asks: who received it in the first place? The machine age forces an earlier question still: was there a real person there? These are different problems. They should not be collapsed into one. --- ## The return of physical presence For decades, digitisation followed a clear direction. Remove the branch. Remove the counter. Remove the paper. Remove the appointment. Remove the human intermediary. Make everything remote. Usually, this was progress. Physical processes impose friction. They require travel, opening hours, staff, buildings and time. Digital services made participation dramatically easier. Artificial intelligence does not reverse those benefits. But it changes the security value of something we spent decades eliminating: **physical presence**. A person standing at a government counter with an identity document creates a very different trust event from a person appearing through an uncontrolled camera connected to an uncontrolled device over the internet. Neither is infallible. Documents can be forged. Officials can make mistakes. People can be coerced. Physical systems can be corrupted. The attack economics are different. Creating ten thousand synthetic video identities can potentially become a software problem. Sending ten thousand distinct human beings to ten thousand physical identity checks remains a physical problem. The physical counter therefore need not be understood as a failure of digitalisation. It can be understood as a **trust anchor**: a deliberately scarce physical event from which many convenient digital interactions can subsequently derive their trust. The objective should not be to make everything physical. It should be to understand where physical reality creates disproportionately valuable assurance. --- ## A physical root, a digital life This leads to a different architecture. Establish strong trust rarely. Use it digitally many times. A person might establish their identity once through a high-assurance process: perhaps at a government office, perhaps through a trusted supervised facility, perhaps through another mechanism capable of providing equivalent assurance. From that event, cryptographically verifiable credentials can be issued. Afterwards, the person should not need to repeatedly reveal their complete identity. Quite the opposite. A strong root of trust should allow us to disclose **less**. A bar does not need my birthday. It needs to know: **OVER 18**. A service restricted to residents may not need my name. It may need: **ELIGIBLE RESIDENT**. A platform trying to control industrial-scale bot creation may not need to know anything about me. It may need: **VERIFIED PERSON**. An election system does not need to attach my identity to my vote. It needs to establish: **ELIGIBLE PERSON + ONE PARTICIPATION**. **Proof of humanity should not require surrendering identity.** A well-designed digital identity infrastructure could make anonymity stronger rather than weaker. It could allow trustworthy claims without requiring universal identification. --- ## Presence is not personhood Even the phrase *proof of human* hides several different questions. They need to be separated. **Presence** asks whether a human is participating in this interaction right now. **Personhood** asks whether there is a distinct real person behind this credential. **Identity** asks which person it is. **Authority** asks what this person is authorised to do. These properties are related, but they are not interchangeable. A system might require personhood without identity. Another might require identity without continuous human presence. And increasingly, a system may need authority without either — because the actor performing the task may not be human at all. --- ## The agent changes the question again The goal cannot be to build an internet in which only humans are allowed to act. We are deliberately building the opposite. AI agents will increasingly communicate, search, negotiate, purchase, schedule, analyse and transact on our behalf. That can be useful. It can also be entirely compatible with [human sovereignty](/field-notes/human-sovereignty). Sovereignty does not require personally performing every action. It requires retaining authority over what may be delegated. A sovereign person should be able to say: this machine may act for me, for this purpose, within these limits, until this time — and I can revoke that authority. The same test that [The Exit Test](/field-notes/the-exit-test) applies to organisations now applies to the person: control only matters when it can become action, including withdrawal. The important distinction of the next internet may therefore not be `HUMAN / BOT`. It may look more like: `VERIFIED HUMAN` `INDEPENDENT MACHINE` `MACHINE ACTING FOR VERIFIED HUMAN` `MACHINE ACTING FOR VERIFIED ORGANISATION` These are very different actors. Our current interfaces rarely distinguish them. Soon they may have to. --- ## A machine should not need to pretend to be me Today, much automation works through impersonation. A program uses my credentials. It opens my browser session. It calls a service using a token issued to me. To the receiving system, the distinction between *me* and *software operating as me* can disappear. That model becomes increasingly dangerous as agents gain capability. A better architecture makes delegation explicit. The service should be able to establish that there is a verified person; that this agent is acting on that person's behalf; that the person granted a defined authority; that the authority has boundaries; that the delegation can expire; and that the person can revoke it. This preserves a chain of authority without requiring the machine to masquerade as the human. It also preserves something essential: **accountability**. Not every machine action should be attributed directly to a person as though they personally performed it. But legitimate machine authority should ultimately have a traceable origin. The question becomes not merely *who did this?* but *under whose authority was this done?* --- ## Delegation without surrender This connects digital identity directly to human sovereignty. Human sovereignty in the machine age cannot mean excluding machines from human affairs. That possibility is already disappearing. Nor can it mean requiring a human confirmation before every machine action. That would eliminate much of the value of delegation. The objective must instead be **delegation without surrender**. Machines can receive authority without acquiring sovereignty. They can act without becoming the source of purpose. They can execute without becoming the owner of the decision. And they can represent us without becoming indistinguishable from us. That requires architecture: identity, delegation, boundaries, attribution, observability, revocation, recovery. The same principles that preserve human agency inside an AI system begin to appear at the level of the internet itself. --- ## The privacy trap There is an obvious danger. A world concerned about bots could respond by demanding universal identification: no anonymity, government identity for every account, a permanent identifier attached to every interaction, every message traceable, every transaction linkable, every digital action attributable to a legal person. That would solve one trust problem by creating a much larger sovereignty problem. Anonymity is not a defect of the internet. It protects whistleblowers, political dissidents, vulnerable people, children, journalists, researchers and ordinary citizens who simply do not want every part of their lives joined into a permanent profile. We should not have to choose between an internet full of synthetic actors and an internet of universal surveillance. That is why the distinction between **identity** and **personhood** matters. Sometimes a system needs to know who I am. Often it does not. It merely needs to know that I am a legitimate participant. A privacy-preserving personhood credential could establish that a real person exists behind this interaction without establishing: **this is Thierry Gilgen**. That difference may become one of the most important design choices of the machine age. --- ## Scarcity returns to the internet There is another reason personhood matters. Humans are scarce. Machines are not. A human being has twenty-four hours in a day. Human attention, judgement and participation are naturally limited. Machine activity can scale with compute. This changes systems whose legitimacy depends implicitly on human scarcity: ratings, petitions, comments, reputation, marketplaces, social networks, community moderation, public consultations, voting — even attention itself. If one person can cheaply create ten thousand convincing machine participants, the problem is not simply that some content is artificial. The relationship between **one participant and one source of agency** has broken. I have written elsewhere about the scramble for undigitised human material once the open web fills with model output ([The Last Human Corpus](/field-notes/the-last-human-corpus)). The same scarcity appears here from the other direction. The corpus problem is that original human data is finite. The participation problem is that original human actors are finite too. Proof of personhood can restore certain forms of scarcity without necessarily revealing identity. Not everywhere. Not for everything. But where the distinction matters. The future internet may therefore need something it never formally possessed: a privacy-preserving way of representing **one human**. --- ## The Swiss warning This is why the delay of a national electronic identity is more interesting than an implementation problem. Switzerland encountered a small preview of a much larger transition. The challenge is not that artificial intelligence has made digital identity impossible. It has not. Remote identity proofing will improve. Document chips provide stronger cryptographic evidence. Liveness detection will evolve. Devices can become more trustworthy. Hybrid processes can combine different assurance mechanisms. The adversary evolves too. We should stop treating physical verification as an embarrassing residue waiting to be eliminated by better software. Sometimes the physical world is precisely where digital trust should begin. The design question is not how we digitise every step. It is where trust should originate, and how we can preserve it afterwards with the least possible disclosure and friction. Those are very different ambitions. --- ## The human root of authority There is a broader architecture emerging. A physical or otherwise high-assurance process establishes a person. A digital credential represents that trust. Privacy-preserving proofs reveal only what a particular interaction requires. Delegation allows machines to receive bounded authority. Cryptography preserves the chain. Revocation allows the human to withdraw it. Recovery prevents the loss of a device from becoming the loss of identity. Throughout the system, one distinction remains visible: where human authority begins and where machine execution takes over. That distinction will matter increasingly as agents communicate primarily with other agents. A future transaction might contain no human interaction at all. My agent negotiates with an insurer's agent. My procurement agent orders from a supplier's agent. My scheduling agent negotiates with another person's scheduling agent. The humans may never see the individual messages. That does not necessarily reduce human sovereignty. But only if the machines remain inside authority structures whose origin and boundaries are humanly determined. The human does not need to remain in every loop. **The human must remain at the root of legitimate authority.** --- ## Proof of human The internet was built during a period in which humanity was cheap to assume. That period is ending. We will increasingly interact with entities whose language, faces, voices and behaviour provide little evidence about what they are. Trying to preserve the old distinction with ever harder CAPTCHAs or ever more invasive surveillance will not be enough. We need a richer trust model: one capable of proving personhood without always revealing identity; of distinguishing machines without excluding them; of allowing agents to act without allowing them to impersonate their principals; of tracing authority without tracing every human action; and of using physical reality where physical reality provides something valuable. Human sovereignty requires more than keeping machines under human control. It requires that **human authority remains recognisable in a world increasingly populated by machines.** The challenge is therefore not to prove that every action was performed by a human. Many will not be. The challenge is to preserve the distinction between human purpose and machine execution — and to make the relationship between them trustworthy. For most of its history, the internet could assume that its users were people. **The next internet will have to prove it.** --- ## Sources 1. **Federal Office of Justice — New timeline for the introduction of the e-ID and the trust infrastructure, 25/30 June 2026** Official notice postponing e-ID introduction. Named reasons: additional AI challenges to the online issuance process, including malware on end-user devices and improved detection of deepfakes. Trust infrastructure still expected in the first half of 2027; no new e-ID date announced. [eid.admin.ch](https://www.eid.admin.ch/en/20260625-neuer-zeitplan-fuer-die-einfuehrung-der-e-id-und-der-vertrauensinfrastruktur-en) ## Connected reading - [Digital sovereignty is not nationalism](https://www.thierry-gilgen-ict.ch/field-notes/digital-sovereignty-is-not-nationalism) - [Human Sovereignty in the Age of Machine Intelligence](https://www.thierry-gilgen-ict.ch/field-notes/human-sovereignty-in-the-age-of-machine-intelligence) - [The Exit Test](https://www.thierry-gilgen-ict.ch/field-notes/the-exit-test) - [The Last Human Corpus](https://www.thierry-gilgen-ict.ch/field-notes/the-last-human-corpus) - [Human Sovereignty](https://www.thierry-gilgen-ict.ch/field-notes/human-sovereignty)