News & Public Record — Essay

The Presence Thesis

Why Physical Verification Is the Last Defence Against AI-Generated Fraud

Rodric David · Founder & CEO, Amplivy · July 2026

In December 2025, I published “The End of Inference” — an argument that the digital economy was experiencing the collapse of presence as a trustworthy primitive. That inference was no longer sufficient. That a verified participation infrastructure was the missing architectural layer.

In the six months since, every verification layer above the physical has been publicly, spectacularly, and irreversibly compromised. This paper documents what has happened, explains what is not being measured, and makes the case that the digital economy’s most urgent need is not better fraud detection. It is the restoration of trust itself.

1. The Trust Crisis

This is not a paper about fraud. Fraud is a symptom. This paper is about trust — the foundational assumption that makes the digital economy possible.

For all of human history, identity and presence were a single fact. You trusted the face in front of you because it was physically there — person and place, who and where, verified together in the same moment. The internet broke that co-truth. It made identity portable — who you are became data that travels — and in doing so it severed identity from presence entirely. Identity is who you are. Presence is whether you, and not someone holding your credentials, are actually there. The digital economy kept the first and simply gave up on the second.

Consider the most ordinary of examples. You check into a hotel. The front desk checks your identity — carefully, even copying your documents. Then it hands you a key. For the rest of the week, no one checks again. The door does not ask who is holding the key. If the key is stolen, the door opens for the thief as readily as it opened for you. Every digital system you use works exactly like that door: identity checked once, a credential issued, and from then on it is the credential — not the human — that is verified. Who a key may admit — one guest, a family — is a decision the hotel can make. The point is that today, the door doesn’t ask at all.

Every digital interaction — every payment, every phone call, every government service, every job interview, every advertisement, every message — depends on a single question that no current technology can definitively answer: is a real human being, acting voluntarily, actually present right now?

The answer today is always a guess. When you approve a payment, your bank does not know you are there. It knows that a device registered to your account, using credentials associated with your identity, sent an instruction. It infers that the instruction came from you. When you answer a phone call, the caller has no verified proof that a human answered. When a business sends you a message, it has no way to confirm that you, as a physical person, actually received it.

For most of the internet era, this was a manageable limitation. The gap between inference and certainty was exploited by criminals, but at a human scale. Fraud required human labour. Scam operations required real people to make calls, conduct conversations, forge documents by hand. The economics of that constraint limited the problem.

Artificial intelligence has removed that constraint entirely.

For hundreds of thousands of years, humans evolved to trust what they could see and hear. If a voice sounded like your grandchild, your brain interpreted that as your grandchild. If a face on a video call looked like your colleague, your brain interpreted that as your colleague. If a document looked like a government-issued ID, your brain interpreted that as authentic.

Artificial intelligence has broken this evolutionary contract. The synthetic voice is indistinguishable from the real one. The deepfake face is indistinguishable from the live person. The AI-generated document is indistinguishable from the genuine article. The biological trust instincts that evolved over hundreds of millennia are no longer reliable guides to digital reality. The erosion of that basic certainty is not a cybersecurity problem. It is a civilisational one.

2. The Acceleration

This is not a future risk. This is the present. The acceleration of AI-generated fraud in the past twelve months has outpaced every projection made by the institutions responsible for preventing it. In July 2026, Cloudflare confirmed that, for the first time in the internet’s history, most internet traffic is not human.

Every one of these numbers is from the past twelve months. Every one represents a rate of change that was not predicted by the institutions responsible for preventing fraud. The digital economy is not facing a future threat. It is losing a war it has not yet recognised it is in.

3. The Dark Figure: What You Haven’t Been Told

The numbers above are alarming. They are also catastrophically incomplete.

Law enforcement agencies estimate that only 2 to 6.7 per cent of fraud victims report their losses. The FTC itself estimates that, adjusted for underreporting, actual US consumer fraud losses in 2024 ran as high as $195.9 billion — against roughly $12.8 billion reported.

The Association of Certified Fraud Examiners (ACFE), in its 2024 Report to the Nations — the largest global study of occupational fraud — estimates that the typical organisation loses 5% of its annual revenue to fraud, whether detected or undetected. Applied to estimated gross world product, 5% translates to roughly $5 trillion in global losses each year — more than the entire GDP of Japan. The US Government Accountability Office (GAO) estimates that the US federal government alone loses between $233 billion and $521 billion annually to fraud.

The median fraud scheme runs for 12 months before it is detected. The median loss runs at $9,900 per month — accumulating silently before anyone knows it is happening. Synthetic identity fraud — where the identity itself is fabricated — is typically cultivated for up to 18 months before the strike, and there is no real person to notice that an identity has been stolen.

Interpol reported $442 billion in global fraud losses for 2025 and called the figure “an underestimate.” The Economist published a briefing in February 2025 titled “Online Scams May Already Be as Big a Scourge as Illegal Drugs.” The Global Anti-Scam Alliance surveyed 46,000 adults across 42 markets and found that 57% had encountered a scam in the past year. UK Finance reported £1.28 billion stolen from British consumers in 2025.

The fraud you read about in headlines is the visible tip of an iceberg. Below the surface: hundreds of billions in unreported consumer losses, trillions in undetected organisational fraud, and hundreds of billions stolen annually from governments. AI is making every category of fraud cheaper to commit, harder to detect, and more profitable to scale. The gap between what is happening and what anyone knows about is widening every quarter.

FBI IC3 Annual Report 2025  |  ACFE Report to the Nations 2024  |  Interpol, March 2026  |  Aspen Institute Fraud Analysis 2025

4. Real People. Real Losses. Real Harm.

The statistics represent millions of individual victims. These are some of their stories — across every category of the digital economy.

The Grandmother

Christine, 86, of South Philadelphia, received a phone call from someone she was certain was her granddaughter. The voice was crying, saying she had been in a car accident. A man identifying himself as an attorney demanded money to secure her granddaughter’s release. Christine paid $6,000 in cash. The scammers collected it from her home. The voice was an AI clone generated from seconds of audio scraped from social media. Christine told NBC: “Another person’s voice, I would have picked up on. But I know this from a baby, so I know her voice.”

NBC Philadelphia  |  Journal of Accountancy, April 2026

The Kidnapping That Never Happened

Jennifer DeStefano of Arizona heard her teenage daughter screaming and crying on a phone call. A man demanded $1 million in ransom. DeStefano was in a state of panic for several minutes before confirming her daughter was safe at home. The voice was an AI clone. DeStefano testified before Congress, calling for legislation to regulate the technology.

SafeWise

The Woman Who Lost Her Home

A woman in Los Angeles was targeted by a romance scam using deepfake video of actor Steve Burton. She believed she was in a genuine relationship. She sent $81,000 in cash, gift cards, and transfers. When the money ran out, she was pressured into selling her home. By the time her family intervened, the home was gone and her life savings had been extracted. Her daughter said: “Conservatorship is the only way. Power of attorney may not be enough.”

ABC7 Los Angeles  |  KTLA, August 2025

The Video Conference Where Nobody Was Real

In January 2024, a finance employee at Arup, a multinational engineering firm, joined a video conference with what he believed were his company’s UK-based Chief Financial Officer and several colleagues. The CFO instructed him to make confidential transfers. Over the course of the call, the employee authorised 15 wire transfers to five bank accounts, totalling US$25.6 million. Every participant on the call was a deepfake. Every face was AI-generated from publicly available video. The employee was the only real human in the meeting. Arup’s CIO later told the World Economic Forum he created a deepfake of himself in 45 minutes with open-source software.

CNN  |  Fortune  |  World Economic Forum

The Call Even Ferrari Almost Believed

In July 2024, an executive at Ferrari received WhatsApp messages and then a live phone call from what sounded exactly like Chief Executive Benedetto Vigna — the voice, the accent, the mannerisms — discussing a confidential acquisition requiring an urgent currency transaction. The voice was an AI clone. The executive grew suspicious and asked a question only the real Vigna could answer: the title of the book he had recommended days earlier. The line went dead. Ferrari lost nothing — because one human happened to ask the right question at the right moment. The deception failed on alertness and luck, not on any system.

Bloomberg  |  Fortune, July 2024

The Bank That Said Yes 8,065 Times

Between January and August 2025, a single financial institution recorded 8,065 attempts to bypass its digital KYC liveness checks for loan applications. The attacks used AI-generated deepfake images injected directly into the biometric video stream. Cost per attack: $10 to $50 for a Deepfake-as-a-Service image. A ready-to-use synthetic identity: up to $15. The World Economic Forum’s Cybercrime Atlas, published January 2026, tested 17 face-swapping tools and 8 camera injection tools against standard bank-grade biometric verification. The finding: most tools bypassed standard biometric checks. When a deepfake bypasses verification, the audit trail shows a clean pass — the system believed it was real.

Group-IB, via Biometric Update, January 2026  |  World Economic Forum, January 2026

The Customer Who Didn’t Exist — 46 Times

In March 2026, an Amsterdam court heard how a single man had defeated ABN AMRO’s digital identity verification 46 times. Using deepfake images built on fraudulently obtained identity documents, he passed the bank’s selfie-to-ID checks again and again, opening 46 accounts in other people’s names that were used to launder the proceeds of online fraud. The verification system recorded 46 clean passes. Not one of the customers was real.

DutchNews.nl  |  Amsterdam District Court, March 2026

The Phone Numbers That Weren’t Theirs

Michael Terpin, a crypto investor, lost approximately $24 million when a teenager paid an AT&T retail employee to transfer Terpin’s phone number to a device the attackers controlled — a SIM swap. On the day FTX collapsed in November 2022, a SIM-swap crew used a fake ID at an AT&T store to take over one employee’s phone number and drained more than $400 million from the exchange; the woman who presented the fake ID was paid $2,500. In January 2024, the US Securities and Exchange Commission itself was compromised: a 25-year-old with a portable ID printer took over the phone number behind the SEC’s official X account, and a fake announcement approving bitcoin ETFs moved the market before the agency could correct it. In March 2025, T-Mobile was ordered to pay $33 million to a customer whose cryptocurrency was drained after a SIM swap. The FBI logged 1,611 SIM-swap complaints and more than $68 million in losses in 2021 — a fivefold jump in a single year — and in the United Kingdom, SIM-swap cases rose 1,055% in 2024. A Princeton University study found that roughly 80% of fraudulent SIM-swap attempts against prepaid accounts succeeded. The phone number — the credential billions of people use for two-factor authentication — can be stolen with a phone call.

Terpin v. AT&T  |  US DOJ  |  Krebs on Security  |  FBI IC3  |  Cifas Fraudscape 2025  |  Princeton (SOUPS 2020)

The Government Services That Trusted Documents

In Australia, scammers exploited the myGov identity system to steal $557 million from taxpayers over two years. They used personal information — much of it suspected to have come from the Optus and Medibank data breaches — to create fake myGov accounts, link them to victims’ Medicare and Centrelink accounts, submit false tax returns, and redirect refunds to scammer bank accounts. One couple lost $800,000. As SBS reported, the Commonwealth Ombudsman’s review found the exploit operated as a “side entrance” that existing security measures did not prevent.

In the United States, a Pakistani-based criminal organisation used artificial intelligence to generate fake audio recordings of Medicare beneficiaries consenting to receive medical products. The AI-generated voices were sold to laboratories and equipment companies, which used them to submit $703 million in false claims to Medicare. The patients were never present. The consent was never given. The voices were synthetic. The government’s verification system accepted it.

SBS News / AU Ombudsman  |  DOJ Healthcare Fraud Takedown, June 2025

The Fake IDs That Cost Nothing

In May 2026, The Atlantic reported that OpenAI’s ChatGPT image generation model can create photorealistic fake government-issued IDs, receipts, and medical prescriptions. The cost: a standard subscription. The skill required: none. The US Treasury’s FinCEN issued a formal alert warning financial institutions that criminals are using generative AI to create fake documents to circumvent customer identification controls.

The Atlantic, May 2026

The Employee Who Didn’t Exist

In July 2024, KnowBe4 — one of the world’s largest cybersecurity training companies — hired a North Korean state-sponsored operative as a remote software engineer. The applicant used AI to alter a stock photo, combined it with a stolen US identity, and passed four video interviews and a full background check. He was caught on his first day installing malware. The US Department of Justice has documented 309 companies that unknowingly hired NK operatives. In June 2025, DOJ searched 29 laptop farms across 16 states. One operative stole $900,000 in cryptocurrency from his employer after being discovered. Palo Alto Networks demonstrated that a convincing deepfake identity for job interviews can be created in 70 minutes on a five-year-old computer.

KnowBe4  |  Axios  |  Palo Alto Networks Unit 42

The Platform That Profits From the Fraud

Internal Meta documents revealed the company projected about 10% of its 2024 revenue — approximately $16 billion — from scam and prohibited advertising. Meta served 15 billion “higher risk” scam ads per day. When its systems identified an advertiser with a 95% probability of running scams, it charged them more rather than banning them. An internal revenue guardrail capped enforcement at 0.15% of revenue. Meta acknowledged anticipated fines were “much smaller” than scam ad revenue. On 21 April 2026, the Consumer Federation of America filed a class action.

Reuters Special Report, November 2025  |  Consumer Federation of America v. Meta, April 2026

The Sheriff and the Cosmetologist

In April 2026, Sheriff Ronny Dodson of Brewster County, Texas, discovered someone had taken his YouTube videos and created a viral deepfake of him endorsing a health supplement. Karen Flowers, a Virginia cosmetologist with a substantial YouTube following, found her image had been deepfaked to sell life insurance. These are not celebrities. A rural sheriff. A cosmetologist. If AI can deepfake them convincingly, it can deepfake anyone. The barrier is no longer fame. It is simply having a few seconds of audio or video online. Which, in 2026, is everyone.

InvestigateTV, April 2026

5. Every Verification Layer Above the Physical Has Been Compromised

The digital economy was built on the assumption that identity could be established through something you know (a password), something you have (a token), or something you are (a biometric). Artificial intelligence has broken all three.

Documents. ChatGPT produces photorealistic fake IDs. FinCEN has issued formal warnings. Synthetic identity document fraud has increased over 300% in the United States. The document-custody model — photograph the ID, run OCR, check a database — is now verifying the quality of the forgery, not the authenticity of the person.

Faces. Real-time face-swapping software runs during live video calls. The verification system sees the victim’s face making natural movements. The face is AI-generated. Gartner: 30% of enterprises will no longer consider face biometrics reliable as a sole factor by 2026. One bank recorded 8,065 deepfake attacks on its liveness checks in eight months. iProov recorded a 1,151% surge in injection attacks on iOS devices in the second half of 2025, and Entrust reports that one in five biometric fraud attempts is now a deepfake.

Voices. A voice clone can be generated from three seconds of audio. It reproduces emotional inflection — fear, urgency, tears. One in four Americans has received an AI deepfake voice call. Among those who received AI voice-clone messages, 77 per cent lost money.

Video. The Arup case proved an entire video conference — multiple participants, faces, voices, synchronised movements — can be fabricated in real time. The employee suspected phishing. Then he joined the call — and the familiar faces and voices dissolved his doubt.

Credentials. SIM swaps defeat phone-based two-factor authentication — in controlled testing, roughly 80% of fraudulent attempts succeeded. Stolen credentials defeat passwords. Real-time phishing proxies intercept hardware token sessions. Each layer was added to compensate for the failure of the layer below it. Each has been compromised by the same acceleration.

The question is no longer whether these layers can be improved. The question is whether there exists a layer that AI cannot reach.

6. The Last Layer Standing

There is one thing artificial intelligence cannot forge: a real human being, physically present. Every other compromised layer — documents, faces, voices, video, credentials — is information, and AI has industrialised the falsification of information. Physical presence is not information. It is a fact about the physical world, and facts about the physical world cannot be mass-produced by software.

In 2017 — before generative AI existed — Amplivy patented the architecture that proves that fact deterministically. It does not ask whether something looks right. It establishes whether a real human is actually there, with certainty rather than probability. How it does so is the subject of the patent estate, not of this essay. What matters here is that the capability exists, that it is patented, and that it is not subject to the arms race that is consuming every probabilistic defence.

7. Events, Not Identities

The most important reframe in understanding the solution is this: the digital economy does not ultimately depend on identity. It depends on events.

Consider what actually needs to be true for a payment to be legitimate. The question is not, abstractly, “who is this person?” The question is: did a real human being, right now, at this moment, voluntarily initiate this specific transaction? The event is what matters. The identity is relevant only insofar as it establishes who performed the event.

None of this diminishes identity. Knowing who someone is remains essential, and the systems that verify identity perform a necessary function. But identity answers only half the question. It establishes who — it cannot establish that the person is actually there at the moment that matters, rather than someone holding their credentials, wearing their face, or speaking in their voice. Presence is the other half. VPI does not check identity, and does not replace the systems that do. It is the companion layer identity has always needed: additive infrastructure beneath them, restoring the half of the truth the internet gave up.

The digital economy runs on tens of trillions of such events every year. A person authorises a payment. A person answers a call. A citizen accesses a government service. A person views an advertisement. An employee accesses a proprietary system. An employee downloads a file, commits code, queries a database. Every one of these events is currently assumed or inferred. None is verified.

Verified Presence Infrastructure (VPI) proposes that they can instead be deterministically verified. At the moment an event occurs, VPI produces a cryptographic proof of presence — tamper-evident confirmation that a real human being was present and in voluntary control during the event window. The proof is delivered to the party that needs it. No profile is built. No history is stored. No surveillance is conducted.

The output is proof, not a profile. Trust is established at the moment it matters, then released.

The examples in this paper address fraud — the most visible and urgent application. But the architectural principle extends to every consequential digital instruction: every database access, every system login, every file transfer, every administrative action in every enterprise, government, and institution in the world. The addressable surface for verified presence is not a sector. It is the digital economy itself.

8. What VPI Does: It Breaks Scale

VPI is not the solution to all fraud. It does not prevent a dishonest employee from stealing while physically present at their desk. It does not cure malicious intent by an individual acting alone, in person, one transaction at a time. That is the person with a gun. VPI does not stop them.

What VPI stops is everything that makes fraud a trillion-dollar global industry: the automation, the remote operation, and the industrial scale.

Every case documented in this paper succeeded because the criminal never had to be physically present for any individual fraudulent act. One call centre in Pakistan generated AI consent recordings for thousands of Medicare patients. One criminal organisation in Russia submitted $10.6 billion in fake claims using a million stolen identities from a room full of computers. One SIM farm operated thousands of “subscribers” from 94 locations across 17 countries. One North Korean operative interviewed for the same job dozens of times using different synthetic faces. Meta served 15 billion scam ads per day generated by bot-driven purchasing.

VPI makes all of it impossible. If every consequential action requires deterministic proof that a real human is present, you cannot automate physical presence. You cannot scale it. You cannot operate it from another country. Every fraudulent transaction now requires a real human body at a real location — and one body can only be in one place at one time.

VPI does not make fraud impossible. It makes fraud unscalable. And unscalable fraud is unprofitable fraud. When the economics of industrial-scale criminal enterprise collapse — when every fraud requires a body in a room — fraud stops being a technology business and goes back to being a street crime. And street crime does not scale to $442 billion.

9. What Would Have Happened

For each case in this paper, consider what would have occurred if VPI had been deployed.

Christine ($6,000). The call arrives carrying no proof that a real, verified person is behind the voice. The AI clone is perfect; the proof does not exist. Christine hangs up. She keeps her $6,000. She is never afraid.

The LA woman ($81,000 + home). The romantic partner can produce no proof that a real, verified person is behind the face on the screen. The deepfake looks real; the proof does not exist. She keeps her savings. She keeps her home.

Arup ($25.6 million). The instruction to transfer arrives without deterministic proof that the real CFO is present and issuing it. The 15 transfers are never authorised. Arup keeps its $25.6 million.

The bank (8,065 deepfake attacks). Each loan application requires deterministic proof of a present, real applicant. A synthetic face cannot supply it. Eight thousand attacks produce zero successful applications.

Medicare ($703 million AI voice fraud). Each patient consent requires deterministic proof that the patient was present and consenting. A synthetic voice cannot supply it. The $703 million in false claims are never submitted.

myGov ($557 million). Linking a Medicare or Centrelink account requires deterministic proof of the account holder’s presence. Stolen documents cannot supply it. The “side entrance” is architecturally closed. The $557 million stays with the taxpayers.

KnowBe4 (NK operative). The interview requires deterministic proof of a present, verified candidate. The synthetic identity cannot supply it. The application is flagged. The spy is never hired.

Meta (15 billion scam ads per day). Each advertisement trading on a person’s identity requires that person’s verified authorisation. The scam advertiser cannot produce it. The 15 billion daily scam ads cannot be placed. Andrew Forrest’s face is never used to sell crypto without his authorisation.

10. The Cost Beyond Money

Every report on fraud quantifies dollars lost. None of them quantifies trust destroyed.

In February 2026, Nikita Bier, the Head of Product at X, predicted that within 90 days iMessage, Gmail, and phone calls would “no longer be usable in any functional sense” due to AI-powered automation. “And we will have no way to stop it,” he wrote. He cited OpenClaw, a platform enabling on-device AI agents that can send texts, draft emails, and automate outreach at negligible cost. The barriers that once limited spam operations — technical knowledge, server infrastructure, capital — have been removed. As of this writing, Bier’s 90-day deadline has arrived. The channels he described are not dead. They are dying of distrust.

McAfee’s 2026 State of the Scamiverse report quantified the cost. Americans now spend 114 hours per year — nearly three full working weeks — trying to determine whether the messages, calls, and content they encounter are real or fake. That number is up from 94 hours the year before. The average American encounters roughly three deepfake videos per day. Britons receive an average of eight scam messages per day across platforms. One in three lost money. Fifty-nine per cent of people globally know someone who has fallen victim to an online scam. Among 18-to-24-year-olds, that figure is 77%.

The trust tax is not measured in dollars lost to fraud. It is measured in hours — in the time every citizen now spends, every day, questioning the reality of their own digital environment. One hundred and fourteen hours per year of cognitive overhead imposed on every participant in the digital economy, with no resolution in sight.

Consider what the daily experience of being a citizen looks like in 2026. You receive scam calls multiple times a week, sometimes daily. Phishing texts pretending to be your bank, your government, your postal service. Emails pretending to be the tax office, your health insurer, your superannuation fund. Deepfake advertisements of people who never endorsed the products they appear to sell. Your parents receive calls from “grandchildren” in distress. Your company receives invoices from “vendors” that don’t exist.

The cumulative effect is not financial. It is psychological.

People stop answering their phones. Legitimate calls from doctors, banks, and government services go unanswered because every call might be a scam.

People stop trusting legitimate digital communications. Thirty-eight per cent of US adults received a call in the past year impersonating their insurance provider. The result: people distrust real communications from their own insurers and doctors.

Elderly people withdraw from technology. The generation most targeted by voice clone scams is the generation least equipped to distinguish real from synthetic. The practical consequence is isolation — from services, from family, from participation in the digital economy.

Government digital services become less effective. With phishing among the leading attack techniques reported against Commonwealth agencies in Australia, citizens learn to distrust every communication that appears to come from myGov, the ATO, Medicare, or Centrelink. The government’s investment in digital service delivery is undermined by its inability to prove its own communications are real.

Businesses cannot reach their own customers. When every call might be a scam, customers stop picking up. The cost is not just fraud prevented — it is legitimate commerce suppressed.

This is what government should care about most. Not the dollars lost to fraud — though those are staggering. The daily erosion of the willingness of citizens to participate in the digital economy. The slow collapse of trust in institutions, in regulation, in law, and in the digital systems that modern society depends on.

The scam that succeeds costs money. The millions of scam attempts that fail — the calls that go unanswered, the texts that are deleted, the emails that are ignored — cost something more important. They cost trust. And trust, once lost, is the one thing the digital economy cannot function without.

11. The Question

Six months ago, I wrote that the digital economy was experiencing the collapse of presence as a trustworthy primitive. Everything that has happened since — the 1,100% increase in deepfake fraud, the $703 million in AI-generated Medicare consent, the bank attacked 8,065 times by synthetic faces, the $557 million stolen through Australia’s central identity system, the cybersecurity company that hired a North Korean spy, the 15 billion scam ads served daily by the platform that calculated fines are cheaper than compliance — confirms that the collapse is real, accelerating, and structural.

The verification layers the world relies on — documents, faces, voices, video, credentials — have all been compromised by AI that is cheaper, faster, and more convincing than the defences designed to stop it. The only verification primitive that survives is physical presence — the one layer that AI cannot reach.

The infrastructure to verify presence at the moment it matters — and release it immediately afterward — exists. It is patent-protected. It builds no profiles. It conducts no surveillance. It produces proof, not profiles. It operates across devices, networks, and institutions. And it becomes more necessary, not less, with every advance in AI capability.

The question is not whether presence will be verified in the future. It will be. The question — as I wrote in December 2025 — is how, by whom, and under what principles.

I believe it should be verified through infrastructure that is consent-based, non-surveillance, deterministic, and sovereign. Infrastructure that protects citizens without monitoring them. Infrastructure that restores the trust the digital economy was built on, and that artificial intelligence is systematically destroying.

That infrastructure is Verified Presence Infrastructure.

That is what Amplivy is building.

And the next front is already visible. The payments industry is deploying AI agents that transact on our behalf — Google concedes they “break the fundamental assumption” that a human clicks buy — while every mechanism shipped so far verifies the agent or the account, never the human. That is the subject of the next essay.