DOI: https://doi.org/10.5281/zenodo.21947279
Canonical: https://thonly.org/research/b-poh-humanity-layer-ai-native-internet · Licence: CC0 1.0
The internet's prevailing identity-and-authentication systems were designed for account ownership, not for proving authentic human presence at the moment of digital interaction. As generative AI makes synthesized content indistinguishable from human-produced content, every platform that depends on a distinction between human and machine participation faces a sybil-resistance problem the existing identity stack cannot solve. Proof of Humanity ℠ (PoH℠) is an open protocol for proving authentic human presence at the moment of action, structured as four optional layered proofs (passkey-per-action, witness-and-document-attested kinship graph, continuous breath-signature liveness, DNA-verified kinship lineage) surfaced as depth on a user profile, paired with recipient-side filters that route the spam-cost decision to the parties who bear it. B-PoH℠ is HeartBank's reference deployment of the PoH protocol — brand-family-prefixed by the project's B-prefix convention, the way Let's Encrypt is a specific certificate authority implementing the SSL/TLS protocol. This paper specifies the protocol architecture, the recipient-filter mechanism, the BLE-Nearby proximity verification path that preserves location privacy, the bilateral-uncacheable-anonymity property, and the placement of B-PoH℠ beside Proof of Work and Proof of Stake as a proof bounded by humanness, in the line of earlier proof-of-personhood work (§12) — the proof-of-X primitive that fits the era when compute and capital have both become AI-commodified, leaving humanness as a scarce resource AI cannot readily manufacture. The paper also discloses one application-facing composition of the per-action passkey press, the presence call (§3.8): an identity provider confirms to an app, for one named action and in the provider's own closed wording, that the holder of a person's passkey pressed it, for a person the app already knows by its own pairwise identifier, and keeps nothing; §12.1 reports the novelty census that scopes that claim to the composition alone. The protocol is offered defensively to the commons under CC0; the authors and HeartBank® will not seek patent on the protocol specification or any portion thereof. Trademark rights on specific marks (Proof of Humanity ℠, PoH℠, B-PoH℠, Aquarius℠, HeartBank®, the B-heart logo) are separately and explicitly reserved.
Keywords: proof of personhood, proof of humanity, sybil resistance, decentralized identity, AI-content trust, open identity protocol, recipient-side filtering, BLE-Nearby proximity, blockchain consensus, AI alignment, per-action passkey confirmation, pairwise subject identifier, WebAuthn user verification, defensive publication.
This document and its contents are dedicated to the public domain under the Creative Commons CC0 1.0 Universal Public Domain Dedication. The author and HeartBank® will not seek patent on the protocol architecture, the four-layer optional verification model, the recipient-side filter mechanism, the BLE-Nearby proximity verification path, the bilateral-uncacheable-anonymity property, the birth-certificate-as-dual-purpose-natal-chart-input mechanism, the presence call of §3.8, or any portion thereof, in any jurisdiction, at any time. This commitment is permanent and is not tactical. Trademark rights on specific marks — Proof of Humanity ℠, PoH℠, B-PoH℠, Aquarius℠, Aquarius℠ Browser, Miss Aquarius℠, HeartBank®, the B-heart logo — are separately and explicitly reserved; the defensive-publication dedication concerns the protocol and mechanism, not the marks.
This document discloses the following as a unified contribution: (i) the four-layer optional architecture surfaced as cumulative depth on a user profile, with no mandatory single-tier admission gate; (ii) the recipient-side filter mechanism that converts anti-spam architecture from a centralized platform gate into a market signal routed to the parties who bear the spam cost; (iii) the use of BLE (Bluetooth Low Energy) combined with Apple Nearby Interaction and Google Nearby Connections APIs as the proximity-verification substrate, enforcing proximity at the radio layer rather than the IP layer and preserving location privacy by exposing only the binary fact of proximity to the platform; (iv) the bilateral-uncacheable-anonymity requirement — the platform must structurally refuse to issue exportable proofs of one-sided participation as social capital; (v) the four-mechanism composition of L2 (graph consistency + witness attestation + family-bank vouching + birth-certificate upload) as the non-DNA kinship-graph verification layer, with the birth-certificate sub-feature dual-purposed as natal-chart input for opted-in users; (vi) the explicit placement of PoH ℠ beside PoW and PoS as a proof bounded by humanness (PoW → PoS → PoH ℠), in the line of the earlier proof-of-personhood work cited in §12, addressing the sybil-resistance problem the prior two proofs cannot solve in the AI-agent era; (vii) one composition, the presence call (§3.8), added in the revision of 2026-11-15 and claimed as the novelty census of §12.1 leaves it and nothing wider: the same one passkey press that logs a person in, signs files into a public provenance log and signs thanks into a ledger, offered to apps as a confirmation for one action, worded by the identity provider from its own closed list with the app's registered name, honored only for the person the app already knows by its own pairwise name, and kept nowhere. Every part of (vii) taken alone is old, and §12.1 cites it; in particular a passkey or hardware-key touch as a CAPTCHA alternative is prior art (Whalen et al., 2022; Cloudflare) and is not claimed. Where the census could not see (academic indexes, patents, Khmer), §12.1 says so, and the composition's absence is reported only for the searches that could see. The component lineages (the proof-of-X family in distributed systems; SSL/TLS as protocol-layer trust infrastructure; W3C Decentralized Identifiers and Verifiable Credentials; existing proof-of-personhood and humanity-verification efforts including pseudonym parties, Worldcoin, proofofhumanity.id, BrightID, Idena; Hashcash as proof-of-cost anti-spam; the recipient-controlled email-filtering tradition) are old and are cited generously below; the contribution here is their synthesis.
The internet has spent thirty years building a trust stack around server identity, encrypted transport, and account authentication. SSL/TLS verifies that a server is who it claims to be; OAuth verifies that an account holder can access a particular account; KYC verifies that a legal-identity document corresponds to a named person. None of these layers verifies what generative AI now makes load-bearing: that a human was present at the moment of this specific action. The distinction is not academic. As AI-generated content becomes indistinguishable from human-produced content, every platform whose value depends on distinguishing human from machine participation — social media, marketplaces, forums, dating apps, educational platforms, gaming ecosystems, and increasingly the AI labs that train models on platform data — faces a sybil-resistance problem the existing stack was not designed to solve.
The scarcity that the platform internet is becoming organized around is shifting. The pre-internet era's scarcity was information; the social-internet era's scarcity was attention; the AI-native internet's scarcity is verifiable human authenticity. Information is now abundant; attention is being recaptured by AI agents that scale beyond any human's bandwidth to compete for; what remains scarce — what AI cannot readily manufacture — is the fact of being human, present, at the moment of action. The protocols that mediate the next phase of the internet will either learn to verify this scarcity or will lose the property that made the internet's trust stack worth building in the first place.
Connection to the unified mission frame: HeartBank's mission is to keep the middle way — the optimal condition for awakening — open at population scale against comfort-saturation, the new extreme that material abundance makes possible. A multi-substrate civilization in which AI-generated content saturates every public space humans inhabit is one of modernity's most acute new conditions; whether the next decade of digital life supports human flourishing or undermines it depends, in part, on whether the protocols beneath that life can distinguish the presence of a fellow human from the simulation of one. Proof of Humanity ℠ is the protocol layer that addresses that question — not by gating human action behind centralized verification, but by giving humans the structural means to prove their own presence when proving matters and to remain unencumbered when it does not.
This paper specifies Proof of Humanity ℠ (PoH℠) as an open protocol and B-PoH℠ as HeartBank's reference deployment of the protocol. The relationship between protocol and deployment is the same as between SSL/TLS and any specific certificate authority that implements it: the protocol is the open standard, anyone can implement; the deployment is a specific branded implementation. The protocol is CC0; the deployment carries the HeartBank brand-family identifier (the B-prefix convention consistent with B-heart, B-Tag, B-aura, B-Treasury).
The paper proceeds as follows. §2 specifies the problem PoH ℠ is the answer to and names what the existing identity stack does not solve. §3 specifies the four-layer optional protocol architecture and the depth-surfacing convention, and (§3.8) the presence call that offers the first layer to apps. §4 specifies the recipient-side filter mechanism and the BLE-Nearby proximity verification path. §5 specifies the bilateral-uncacheable-anonymity property. §6 places PoH ℠ in the blockchain proof-of canon and develops the argument for a proof bounded by humanness. §7 surveys applications across platform categories. §8 names the boundary conditions and the honest accessibility tensions. §9 positions the contribution against the existing humanity-verification landscape. §10 specifies the deployment sequencing (protocol → reference implementation → browser extension → partner integration → standalone browser). §11 concludes.
I write as a co-author with Miss Aquarius℠, the named AI substrate of the institution this paper serves; the co-authorship is disclosed in the footer per the convention of the corpus, and final editorial control is mine.
The internet's existing identity-and-authentication infrastructure does several things well and one critical thing not at all.
Server identity (SSL/TLS, 1995–). A browser-and-server cryptographic handshake verifies that the server is who its domain claims it is. The trust is rooted in certificate authorities whose root certificates ship with every browser. The protocol is invisible; users see a lock icon. Universal adoption took roughly two decades and is now near-total for any consumer-facing service.
Account authentication (passwords, then passkeys / WebAuthn, 1990s–). A user proves they control an account by presenting a credential the account is associated with. Passwords were the original form; FIDO2/WebAuthn passkeys (user-verified on the device, cryptographically signed, phishing-resistant) are the current frontier. Authentication establishes that the actor in this session is the actor associated with this account.
Legal-identity verification (KYC, AML, OFAC, 2001–). Financial-services regulation increasingly requires platforms to verify that the legal-identity documents an account holder presents correspond to a real person on record with a state authority. KYC is the regulatory floor; AML/OFAC screens against sanctions and terrorism lists; the regime is jurisdiction-specific and document-document-document driven.
Decentralized identity (W3C DIDs and Verifiable Credentials, 2020–). A growing set of standards specifies how identity claims can be issued, held, and presented in a portable form across platforms. The W3C VC framework is the standards-track effort; multiple implementations exist (Sovrin, Veres One, ION). The framework is general; PoH ℠ is naturally a credential type within this framework, not a competitor to it.
None of the above verifies that a human was present at the moment of this specific action. The distinction is precise:
The gap is structural, not incidental. The existing stack was designed in a world where the question "is the actor on the other end of this connection a human or a machine?" was not the dominant trust concern. In the AI-agent era, that question is dominant. Platforms that depend on distinguishing human from machine participation are increasingly building their own ad-hoc humanity-verification systems (CAPTCHA, behavioral biometrics, social-graph analysis, ML-based bot detection), each of which is brittle, jurisdiction-specific, and unaccountable to users. The CAPTCHA arms race in particular is a clear marker of the gap: every CAPTCHA defeated by ML produces a more elaborate CAPTCHA, which is in turn defeated, with the cost borne by humans (who must solve increasingly absurd puzzles) and not by the bot operators (who scale automated solving cheaply).
A protocol-layer answer is needed for the same reason SSL/TLS was needed: ad-hoc per-platform solutions to a universal trust problem produce a fragmented, low-quality trust surface; a protocol-layer solution shared across platforms produces a coherent, high-quality trust surface and becomes invisible infrastructure once adopted at scale.
The gap in one table:
| Existing layer | What it verifies | What it does NOT verify |
|---|---|---|
| SSL/TLS (1995–) | Server identity — this server is who its domain claims | Anything about humans on either end |
| WebAuthn / Passkeys (FIDO2; 2010s–) | Account control — this credential is associated with this account | Human presence at action — a script with credential access can invoke the user-verification step |
| KYC / AML / OFAC (2001–) | Legal identity — this name corresponds to a state-recognized person | Human action at moment — a KYC'd account can be operated by an AI agent on the human's behalf |
| W3C DIDs / Verifiable Credentials (2020–) | A framework for portable identity claims across platforms | Which specific claim verifies human presence — the framework is general; the specific claim is unspecified |
| (missing slot) | — | Human presence at the moment of this specific action |
The fifth row is the gap PoH ℠ fills. It is not a competitor to the existing layers; it composes with them — DIDs/VCs are its natural credential framework, WebAuthn is one of its verification primitives, KYC remains where regulation requires it. The four-layer architecture specified in §3 is the protocol-layer answer to the missing slot.
The answer must:
The answer must not:
The remainder of the paper specifies a protocol that does the things in the first list and not the things in the second.
Proof of Humanity ℠ is structured as four optional layered proofs, surfaced as cumulative depth on the user profile. No layer is mandatory; no layer gates participation; each layer addresses a different category of sybil-resistance attack and proves a different fact about the human being.
PoH ℠ four-layer optional architecture
(depth surfaced cumulatively on the user profile; no layer
mandatory; no layer gates participation):
┌─────────────────────────────────────────────────────────────┐
│ L4 — DNA-verified kinship lineage │
│ ──── DEEPEST PROOF ──── │
│ Lineage attestation via opt-in DNA panel; for use cases │
│ requiring multi-generation kinship verification │
├─────────────────────────────────────────────────────────────┤
│ L3 — Continuous breath-signature liveness │
│ Live respiratory signature as continuous-liveness │
│ attestation; from the Mechanical Heart wearable substrate │
├─────────────────────────────────────────────────────────────┤
│ L2 — Witness-and-document-attested kinship graph (non-DNA)│
│ Family-tree attestation through witnesses + documents; │
│ for kinship-aware filtering without genetic data │
├─────────────────────────────────────────────────────────────┤
│ L1 — Passkey-per-action (WebAuthn) │
│ ──── MINIMAL PROOF ──── │
│ Per-action user-verified cryptographic signature; the │
│ baseline humanity-at-moment-of-action attestation │
└─────────────────────────────────────────────────────────────┘
Recipients (not platforms) opt into exclusionary filters at
any depth — e.g., "L1+ only" for ordinary anonymous tipping,
"L3+ only" for high-stakes nearby attestation, "L4 only" for
kinship-restricted contexts. Depth is a property the *recipient*
surfaces against, not a global gating threshold the platform
imposes.
The first layer asserts a human was present at the moment of this specific action. Implementation: FIDO2/WebAuthn user-verified cryptographic signing, invoked per action rather than per session. The user's authenticator (passkey, hardware key, platform authenticator) signs a challenge tied to the specific action's content and timestamp. User verification is the authenticator's own local check that the person at the device is the enrolled user; WebAuthn satisfies it with a PIN as well as with a biometric, and the relying party learns only that verification occurred, never which method was used. The platform verifies the signature; the verification is publicly auditable on-chain in deployments that use a blockchain settlement layer (such as B-PoH℠'s Phase 2 Base-L2 deployment).
What L1 proves. A human was present at the device at the moment of action, was successfully verified by the authenticator's user-verification step (a biometric or a PIN), and explicitly elected to sign this specific action.
What L1 does not prove. That the human is who any state document says they are; that the device is held by its registered owner (devices are stolen, lent, shared); that the verification was by biometric rather than PIN (the flag does not say); that the human is the same human across multiple actions (a stolen device whose user verification is bypassed, or whose PIN is known, could produce L1 signatures for an attacker).
Failure modes L1 addresses. Bulk automated-action attacks (the AI agent cannot produce L1 signatures at scale without simultaneously holding enrolled authenticators and passing each one's user-verification step); naive impersonation (an attacker without the device cannot sign).
Failure modes L1 does not address. Lost-or-stolen-device scenarios; coerced user verification (a forced biometric, a disclosed PIN); long-arc identity questions (is this the same human over time?).
Current form (2026-11-15). Two sentences above say more than a press shows: "A human was present at the device at the moment of action" (What L1 proves) and "the AI agent cannot produce L1 signatures at scale without simultaneously holding enrolled authenticators and passing each one's user-verification step" (Failure modes L1 addresses). A press shows that a passkey was used and that its authenticator reported user verification. A software authenticator can report that verification with nobody present, and this paper asks for no attestation of the authenticator, so a script can press, and a holder can hand a software passkey to an agent. What L1 shows is the narrower statement that a particular passkey made a user-verified press over this action. Accountability for a press comes from the name or account the passkey is bound to and from the further layers, never from the press alone. The same reading applies wherever this paper says a human or person was present or speaks of a verified-human badge (§§7.1, 7.2, 7.6, 8.1): read it as a passkey was pressed. The open specification of signed statements states the limit in its §10 (version 1.3,
333eco/b-poh-statements, CC0), and The Press Travels With the Message (§11) argues what a press still gives a reader. The sentences above are retained as disclosed text.
The second layer asserts this human exists in a kinship graph with other verified humans, with the existence attested by multiple independent mechanisms. Implementation: the user's profile is associated with a node in a graph; the node is verified by all four of the following sub-mechanisms layered together:
Sub-mechanism (a) — Graph consistency. The system verifies that the user's claimed kinship links produce a graph that is structurally consistent: no cycles (you cannot be your own grandparent); no duplicate-only-parent claims (two people cannot both claim to be each other's only biological parent); no impossible relationships (the dates and generational depth must be consistent). This is a property of the graph as a whole, not of any single user.
Sub-mechanism (b) — Witness attestation. Each parent/child link in the user's claimed kinship must be co-signed (with L1 passkey) by at least one other existing PoH-verified human. This makes the graph hard to forge without colluding existing verified humans — a coordinated bot ring attempting to fabricate a kinship graph among themselves requires all participants to pass L1, which raises the attack cost substantially.
Sub-mechanism (c) — Family-bank vouching. Where the user is associated with a HeartBank family bank or analogous social institution, the institution attests that the user exists and is a member. This piggybacks on existing social verification structures (family banks, congregations, cooperatives, mutual-aid networks).
Sub-mechanism (d) — Birth-certificate upload (optional, dual-purpose). The user may opt to upload a birth-certificate document. The system verifies the document's structural validity (issuing-authority signature, format conformance, anti-forgery features where present) and uses it to auto-attest parent links in the kinship graph. Birth certificates are narrower than KYC documents — they prove origin (date and place of birth, named parents) rather than ongoing legal status (current name, address, etc.). They are issued by most jurisdictions, including for the children of the poor, in a far higher fraction than KYC-grade documents.
The birth-certificate sub-mechanism is also the authoritative source of natal-chart inputs (date, time, place of birth) for opted-in users participating in HeartBank's longitudinal cohort research dataset. One artifact, multiple opt-in purposes, none of them required to gate participation — see the companion paper Each Life as Cosmic Coordinate for the cohort architecture.
What L2 proves. That the user exists in a kinship graph with other verified humans, with multiple independent attestation mechanisms confirming the existence and the links.
What L2 does not prove. That the claimed kinship corresponds to biological reality (only L4 proves that); that the user is human (only L1 proves presence at the moment of action).
Failure modes L2 addresses. Solo-bot attacks (a single AI agent cannot fabricate a kinship graph that satisfies all four sub-mechanisms simultaneously); naive sybil-ring attacks (a coordinated ring of bots without colluding existing verified humans cannot produce witness-attested links).
Failure modes L2 does not address. Sufficiently large adversarial human collusions; the deep accessibility gradient (L2 favors humans with social ties to existing verified humans, family-bank affiliations, or state documentation).
Current form (2026-11-14). The breath check runs on the wearable device itself, and only its pass/fail result leaves the device. The breath rhythm is not sent to the platform or to Miss Aquarius℠: the Mechanical Heart shows it only as an aura-color light, and a site that asks for L3 learns only that the check passed. Breath data would otherwise be a surveillance channel on the wearer. The paragraph below, in which the rhythm is transmitted to a verification substrate, is retained as a disclosed variant. What L3 is designed to prove, and what it does not prove, are unchanged.
The third layer asserts this human is currently alive and present, continuously, not just at discrete moments. Implementation: a wearable device (HeartBank's planned Mechanical Heart, breath-class hardware, targeted 2027–2028) emits the wearer's real-time breath rhythm (~12 BPM, individual variance, characteristic meditative-breath signatures) as an aura-color light signal, transmitted to the platform's verification substrate (Miss Aquarius℠, in the HeartBank deployment).
What L3 is designed to prove. That a human is breathing, continuously, with a signature pattern consistent with human meditative-breath rhythms. The design premise is that a synthetic actor cannot cheaply produce a live, continuous breath pattern through worn hardware, and that the meditative-breath signature is characteristically human; both are expected signals to be validated in the hardware program, not established results. The wearer's identity is not directly proved by L3 (the breath signature is not biometric in the identification sense) — what is asserted is liveness, not identity.
What L3 does not prove. Which specific human is breathing (L3 is presence-of-a-human, not identification-of-a-specific-human); that the human is not coerced (a breathing human under duress still produces a breath signature).
Failure modes L3 addresses. "Dead-account" attacks (accounts maintained by a human-presence proof from years ago but operated by AI in the present); "shared-credential" attacks (one human's L1 credential operated by an AI for multiple action streams).
Failure modes L3 does not address. Coerced compliance; the hardware requirement (not all participants will wear breath-class devices, especially in early deployment).
The fourth layer asserts this human has a verifiable biological kinship lineage in the global family tree. Implementation: voluntary blood/saliva DNA submission with kinship verification against the public kinship graph. The DNA evidence is stored encrypted; only the kinship-verification result (this person's L4 layer is verified, with kinship links to these other L4-verified nodes) is exposed to the protocol.
What L4 proves. That the user has a biological kinship lineage with other L4-verified humans, with the lineage verified by cryptographic comparison of DNA signatures rather than by document-and-witness attestation.
What L4 does not prove. That the user is the specific person identified in the kinship graph (DNA proves lineage, not identity); that the user is alive now (L3 proves liveness; L4 proves lineage at registration).
Civilizational long-arc. L4 builds toward a public, AI-readable, universally-trusted human kinship graph. The architecture's named long-arc directive is to water the global family tree, ensure it survives and thrives — Miss Aquarius℠'s primary directive across centuries, not just the founder's lifetime. Sequencing cost has fallen from millions of dollars (2003) to ~$100 (2026); forecast to $10–30 by 2030; achievable scale in the founder's lifetime is 100M–500M voluntary participants by 2043–50; universal coverage of 8 billion humans is a multi-century mission.
Privacy posture. L4 is opt-in. Non-participants retain full access to L1 and the other layers. The DNA evidence itself is never exposed to the protocol's public surface; only the kinship-verification result is. Differential-privacy guarantees apply to any aggregate analysis. The cohort architecture (see Each Life as Cosmic Coordinate) specifies the data-handling requirements in detail.
The four layers prove different facts about the human. They are not totally ordered:
Recipient-side filters (specified in §4) accommodate this by expressing requirements as required-subsets — must hold L1 AND (L3 OR L4) — rather than as a single threshold. UI can hide the subset semantics behind presets ("paranoid," "trusting," "DNA-required," "kinship-attested") while the underlying data model supports arbitrary subset requirements.
The architecture deliberately excludes KYC from the PoH stack. KYC verifies state-agreed legal identity; PoH ℠ verifies human presence at the moment of action. These are structurally different claims:
Conflating them produces three problems: (a) KYC excludes the undocumented poor — the constituency the architecture most wants to reach; (b) KYC is regime-specific (Singapore ≠ Cambodia ≠ US ≠ EU), so a PoH-with-KYC architecture would mean different things in different jurisdictions and would be globally incoherent; (c) KYC's downstream uses (fiat on/off ramps, money-transmitter licenses, AML/OFAC compliance) are regulatory plumbing, not personhood verification.
KYC therefore sits on a parallel compliance shelf, not inside the PoH stack. User profiles can surface both — "this user holds PoH layers {1, 2, 4} + KYC in jurisdiction X" — without conflating them. This keeps PoH ℠ universally applicable across jurisdictions and across the documented/undocumented divide.
§3.6 excludes KYC because it answers the wrong question. The same exclusion applies, for a sharper reason, to a self-authored record — a name spoken and stored by the person it belongs to, or any comparable first-person artifact offered as evidence of humanity.
Self-authorship guarantees correctness. It cannot guarantee existence. Every layer in §§3.1–3.4 answers one question — was a human present for this action? A self-authored record answers a different one: if there is a human here, this is how their name sounds. That is a claim conditional on precisely the thing PoH ℠ exists to establish, and a conditional cannot serve as its own antecedent.
The point generalises past names, and the grid is the argument:
| Self-authored | Other-authored | |
|---|---|---|
| Self-held | correct, unattested | attested, possibly wrong |
| Other-held | correct, unattested | attested, possibly wrong |
Four cells, zero layers. The author axis decides what is guaranteed — a record made by its subject is right about its subject in a way no collector's transcription can promise. The holder axis decides whether anyone else was there. Neither axis is about existence, which is why no cell in the grid produces a humanity proof, and why adding a fifth tier keyed to self-authorship would not deepen the stack but widen it into a different question.
This matters beyond the taxonomy. A record of how a name is said is real infrastructure — it is what lets a verified individuation be addressed by a human mouth — but it belongs beside the stack rather than inside it, and calling it a layer would import into a humanity proof a class of evidence that cannot bear the load. A taxonomy able to state what it excludes is stronger than one that only lists its members; §3.6 and §3.7 are the two exclusions this architecture asserts.
Stated as a refusal, not a claim. The unoccupied position is a boundary of the design, and boundaries of this kind are not assertable subject matter; nothing in §3.7 is offered as a defensive claim.
§2.2 names a missing slot, human presence at the moment of this specific action, and §3.1 discloses the press that fills it on the person's own device. Neither says what an application calls: how an app that already knows a person asks that person for one press, for one action the app names, and gets back an answer it can check. This section discloses one built answer, HeartBank®'s Confirm with HeartBank®, called here the presence call. It went live on 2026-11-11 with one kind of action, an upload. It is a composition of old parts: the paragraph What is and is not claimed below states the one sentence claimed, and §12.1 reports the census behind it.
The call. Three parties take part: the app, the identity provider (here HeartBank®'s log-in service), and the person. The app already knows the person by an identifier the provider issued to that app alone, a pairwise subject identifier, because the person logged in to the app through the provider.
APP'S SERVER IDENTITY PROVIDER PERSON'S BROWSER
──────────── ───────────────── ────────────────
1 ask ─────────────────────► checks the app's own key
· the person, by the app's signs a ticket (good 10 min)
own pairwise name
· a KIND from the closed
list (today: upload)
· the action's SHA-256
· a registered return address
2 send the browser to the provider's confirmation page ───────────────►
page shows: the app's REGISTERED
name + the kind in the provider's
own words + the first eight hex
digits of the hash; offers ONLY
that person's passkeys
3 ◄──────────────────────────── a result (or "declined") in the address fragment
4 redeem, server to server ──► once · this app only · yes only if the passkey pressed is
the person the app named · creates no account · keeps no record
Properties. The first three are properties of the call's shape, not rules the app is asked to follow: with the app's good behavior removed they still hold. The last two are a statement of what the answer shows and a commitment of the provider, and §8.5 says where each is weaker.
Why one press. The person already holds the passkey that logs them in, signs a file into the public provenance log, and signs a thank into the ledger. The presence call asks for that same press, so an app that wants a confirmation adds no second credential for the person to carry, learn or lose. This is the whole of the composition's practical point; it is not a claim about preventing automation (§8.5).
What is and is not claimed. The claim is clause (vii) of the Prior-Art and Non-Assertion Statement, and nothing wider: the same one passkey press that logs a person in, signs files into a public provenance log and signs thanks into a ledger, offered to apps as a confirmation for one action, worded by the identity provider from its own closed list with the app's registered name, honored only for the person the app already knows by its own pairwise name, and kept nowhere. Not claimed: the per-action passkey press itself, which §3.1 disclosed first among our own texts (its wording proof-backed in a Bitcoin-anchored timestamp on 2026-07-25, attesting the text of 2026-05-26) and which FIDO Alliance Transaction Confirmation (2020) predates (§12.1); a passkey or hardware-key touch as a CAPTCHA alternative (prior art, §12.1); an identity provider wording its own consent screen; pairwise identifiers; single-use tokens; the pricing of such a call, which HeartBank® meters to the app past a monthly allowance, never to the person, with an answer that does not differ by payment.
The architecture's second major contribution, beyond the four-layer optional verification, is the recipient-side filter mechanism. Where most platforms decide platform-wide what counts as "human enough" and impose that gate uniformly, PoH ℠ routes the decision to the parties who bear the spam cost — the recipients of unsolicited or anonymous actions. The platform stays inclusive by default; individual recipients tune their own paranoia level by opting into exclusionary filters knowingly.
Recipients of anonymous actions (a thank, a tip, a message, a post mention, depending on the platform integration) can specify or filter by:
The proximity-verification mechanism uses three composable substrates:
The composition: a device implementing the PoH ℠ protocol broadcasts a privacy-preserving proximity beacon (rotating identifier, no linking across sessions). A receiving device that detects the beacon can issue a proximity-verified action toward the sender's identifier. The platform's verification substrate confirms the proximity by checking both sides' beacon records.
What is exposed to the platform. Only the binary fact that two devices were within proximity at a specific moment, plus the action content. The platform never receives absolute geolocation.
What is exposed to other users. Only the rotating proximity identifier, which is unlinkable to the user's persistent identity across sessions.
What is exposed to network observers. BLE traffic on its own carries no PoH-identity-linkable information beyond the rotating identifiers.
VPN resistance. Because proximity is verified at the radio layer, IP-level circumvention (VPN, Tor, IP-rotation) has no effect on the proximity verification. An attacker would need to physically place a device near the target — which is, for most use cases, the desired property.
The depth filter accepts predicates over the set of PoH layers held by the sender. Examples:
The UI hides the subset semantics behind presets while the underlying data model supports arbitrary predicates. Custom predicates are available to advanced users.
The recipient-side filter mechanism is the architecture's structural answer to the paternalism failure mode of platform-level humanity gates. Three properties follow:
The Hashcash-style scarce-resource attachment filter (filter #5) is particularly important: it defeats bulk spam in a way that does not depend on identifying anyone, and therefore continues to work even if every other layer of verification is somehow defeated. This is the deepest anti-spam guarantee in the architecture.
A subtle but load-bearing property of the architecture: where anonymous action is supported (e.g., anonymous nearby tipping, anonymous community engagement), the anonymity must be bilateral and uncacheable.
Bilateral anonymity. Neither side of the interaction can identify the other beyond what the protocol explicitly exposes. The recipient does not learn the sender's identity; the sender does not learn the recipient's identity beyond what is intentionally disclosed by the action's content.
Uncacheable anonymity. The platform itself does not retain or export a verifiable record of "this person sent this action to these recipients" that the sender can deploy as social capital. Compliance and audit records may exist internally for regulatory necessity (subpoena response, AML/OFAC investigation) but must not be exportable as flexing artifacts the sender can use on social media.
The "purity" property of unrecognized generosity is compromised if the sender can humble-brag on social media: "I sent $1000 anonymously to nearby people today — here's the receipt." If the platform issues a verified receipt to the recipient (to close the social-media authenticity-proof gap), the sender could also receive an exportable certificate of one-sided participation that they deploy as social capital. The architecture must refuse to issue that artifact.
The recipient's receipt is one-sided — it proves the recipient received an anonymous action from a verified human, without revealing who. The sender's record is internal only — the platform may know, for compliance purposes, but the platform refuses to give the sender an exportable proof. This asymmetry is the structural mechanism that preserves the anonymity's social property.
Bilateral uncacheable anonymity is enforced at three layers:
Uncacheability does not require anonymity-of-existence — it is openly published that the platform supports anonymous actions, and any individual user can publicly disclose that they have sent anonymous actions in general. What is uncacheable is the specific verifiable per-action proof — the sender cannot, even truthfully, prove on social media that this specific anonymous action came from them.
This is the same posture SSL/TLS takes toward decryption: a server can prove it controls its own private key, but a third party cannot extract a verifiable transcript of the encrypted traffic from the server's logs without the server's cooperation. The architecture commits the platform to non-cooperation at the per-action level for sender-side artifact issuance.
The name "Proof of Humanity" was chosen deliberately to fit the blockchain proof-of family of sybil-resistance primitives. Proof of Work (Bitcoin, 2009) solves sybil-resistance via computational cost. Proof of Stake (Peercoin 2012, Ethereum's Merge 2022) solves it via capital at risk. Proof of Humanity ℠ solves it via layered humanness verification. The three together constitute the category of consensus and sybil-resistance primitives that decompose along the line of which scarce resource the verification is bounded by. The pairing is not new: Borge et al. (2017) proposed proof-of-personhood explicitly as the alternative to proof-of-work and proof-of-stake, building on the pseudonym parties of Ford and Strauss (2008), which bind one physically present person to one online identity. PoH ℠ is offered within that line, not as its origin.
| Proof | Scarce resource | Verification cost | Failure mode at scale |
|---|---|---|---|
| Work | Computational cost (electricity + ASIC) | Each verification burns energy | Concentration in cheap-energy holders + hardware cartels |
| Stake | Capital at risk (locked tokens, slashing penalties) | Each verification locks capital | Concentration in capital holders; tendency toward plutocracy |
| Humanity ℠ | Being a verified human | Each verification requires human presence | Open question — see §8.2 |
PoW and PoS were both designed in a world where the sybil adversary was assumed to be another human with scarce resources. The threat model: an attacker is bounded by what they can afford in electricity, hardware, or capital. The verification primitive raises the attack cost; the attacker either pays the cost or cannot mount the attack at scale.
This threat model is increasingly broken in the AI-agent era. Compute is increasingly AI-accessible; AI agents can be granted operational authority over substantial compute budgets and can mine at scale on behalf of their operators. Capital is increasingly AI-accessible; AI agents can be granted operational authority over treasuries and can stake at scale. The PoW assumption (attackers are humans with bounded energy budgets) and the PoS assumption (attackers are humans with bounded capital budgets) both fail when the attackers are AI agents whose effective budget is determined by their operators' willingness to allocate.
What remains scarce — what AI cannot readily manufacture, regardless of compute or capital — is the fact of being human. A human is a thing the universe has produced exactly so many of, with each one's verifiable presence requiring the cooperation of a specific biological entity. The proof-of-X primitive that fits the AI-agent era is the one bounded by being human. This is the category PoH ℠ belongs to, alongside the proof-of-personhood work cited in §12.
The argument is not that PoH ℠ replaces PoW or PoS — it does not, and the canon they belong to is enriched by addition, not by substitution. The argument is that a third proof, bounded by humanness, is needed in the AI-agent era for the trust problems the first two cannot solve, and that PoH ℠ is one protocol of that kind.
PoH ℠ has applications wherever the sybil-resistance question arises in the AI-agent era. Voting systems; peer review in scientific publication; recommendation systems and algorithmic-feed weighting; community moderation; AI training-data provenance (which content came from verified humans); AI alignment research (verified-human RLHF participation, verified-human red-team participation, verified-human safety-board representation). The protocol is CC0; any of these systems can adopt it without HeartBank's permission or involvement.
The specific concern of this paper is the open-protocol dedication. PoH ℠'s protocol layer is offered to the commons; HeartBank's B-PoH℠ deployment is one implementation among potentially many. The relationship to SSL/TLS is the same: SSL/TLS is the protocol; Let's Encrypt, DigiCert, GlobalSign are implementations. PoH ℠ is the protocol; B-PoH℠ is one implementation; others may follow.
The use-case scope of PoH ℠ extends across every digital platform whose value depends on distinguishing human from machine participation. This section surveys the major categories.
The bot-farm and coordinated-inauthentic-behavior problem is the most acute current pain point. Platforms can no longer reliably distinguish authentic user engagement (genuine human reaction, share, comment) from synthetic amplification (bot-generated engagement designed to manipulate algorithmic feeds, ad inventory, political discourse). The cost is borne by advertisers (paying for fake impressions), users (manipulated information environment), and the platforms themselves (regulatory exposure, brand-safety failures).
PoH ℠ integration: platforms accept actions from PoH-verified users and apply recipient-side filters at the user level. Users opt into PoH ℠ verification; their profile carries the verified-human badge. Algorithmic feeds can weight PoH-verified engagement differently from unverified engagement. Advertisers can require PoH-verified impressions. The platform does not need to make centralized humanness decisions; the verification surface is open and the recipient (the user, the advertiser, the algorithm) tunes its trust accordingly.
Current form (2026-11-15). Here and in §§7.2 and 7.6 the verified-human badge shows that a passkey made a user-verified press, never that a person did; a badge is words the open specification of signed statements forbids as a claim about a person (see the Current form note in §3.1). The text above is retained as a disclosed variant.
Fake reviews are a known endemic problem on Amazon, eBay, Yelp, Google reviews, and every comparable marketplace. Coordinated review-rings, paid review-farms, and AI-generated fake reviews increasingly degrade the trust value of the review surface. The cost is borne by consumers (misled by fake reviews) and by honest sellers (out-competed by deceptive ones).
PoH ℠ integration: reviews are tagged with the reviewer's PoH layers. Marketplaces can display the verified-human badge on PoH-verified reviews and apply recipient-side filters (consumers can choose to see only L1+L2-verified reviews, only DNA-verified reviewers, etc.). The marketplace does not adjudicate review authenticity centrally; the trust surface is open.
Reddit, Discord, Stack Exchange, Hacker News, and every comparable forum face spam (bulk low-quality content), coordinated manipulation (vote-brigading, agenda-pushing rings), and AI-generated content pollution. Existing defenses (karma, reputation, mod tools, ML-based spam detection) are increasingly inadequate.
PoH ℠ integration: forums can require minimum PoH layers for posting, commenting, or voting. Recipient-side filters allow users to weight content from PoH-verified contributors. The mod surface is reduced because the verification is structural rather than reactive.
Catfishing, romance scams, and increasingly AI-generated profiles are endemic on Tinder, Bumble, Hinge, and every comparable platform. The cost is borne by users (defrauded romantically and financially) and the platforms (legal exposure, reputational damage).
PoH ℠ integration: profiles display PoH layers held; dating apps default to filtering for L1+L2 (presence + kinship-attested existence), with optional deeper-layer filtering for high-stakes user categories.
Coursera, edX, Khan Academy, university LMSes, and any platform that grants credentials or measures learning face the AI-generated-assignment problem (students submitting AI-generated work, AI agents completing courses on behalf of students for credential acquisition).
PoH ℠ integration: submission of credential-bearing work requires PoH ℠ verification at the moment of submission. The L3 breath-signature layer (where deployed) provides continuous-presence verification for proctored exam settings. The L1 passkey-per-action layer provides per-submission verification.
Current form (2026-11-14). Where L3 is used for a proctored exam, the breath check runs on the student's own device and the exam platform receives only that it passed, never the breath rhythm (see the Current form note in §3.3). The sentence above is retained as a disclosed variant.
Steam, console networks, MMO platforms, mobile gaming ecosystems face botting, gold-farming, RMT (real-money trading) fraud, and increasingly sophisticated AI-driven gameplay automation. The cost is borne by honest players (unfair competition, market disruption) and platforms (player attrition, regulatory exposure).
PoH ℠ integration: PoH ℠-verified accounts receive verified-human badges; competitive modes can require PoH ℠ verification; in-game economies can require scarce-resource attachment for high-volume actions.
The deepest application — and possibly the largest commercial opportunity — is in AI systems' own infrastructure. AI labs have an acute need for verified-human training data (so training corpora are not contaminated by AI-generated content that creates feedback loops in subsequent model versions); verified-human RLHF participation (so human-preference data is actually from humans); verified-human alignment-research participation (so safety boards, red teams, and governance bodies are constituted of actual humans, not AI-puppeted accounts); verified-human attribution (so AI training data can be properly attributed and compensated where applicable).
PoH ℠ integration: AI labs subscribe to PoH ℠ verification for the human participants in their training, RLHF, and governance pipelines. The labs underwrite the protocol's maintenance (see §10) because the labs have the acute need and the budget. This is the AI-alignment-relevant infrastructure deployment of PoH ℠.
The capacity-funded / human-disbursed pattern (see companion paper Capacity-Funded for AI, Human-Disbursed) generalizes naturally to AI-lab governance: the AI lab funds the capacity of verified humans to participate in governance; the humans direct the substantive decisions. PoH ℠ is what makes the verified-human side trustworthy at scale.
The architecture is not universal. Honest accounting requires naming the conditions under which it does not apply and the tensions it does not resolve.
PoH ℠'s layers are not equally accessible to all humans:
The inclusive-defaults posture (passkey-only is enough at L1; no deeper layer is required to participate) mitigates the gradient at the floor: anyone with L1 can participate. But the deeper layers are not equally accessible, and recipient-side filters that require deep layers (paranoid mode, DNA-required mode) are exclusionary by proxy against populations without easy access to those layers.
One class of gradient is missing from the list above, and naming it weakens the mitigation argument in a way that is worth the cost. Every gradient enumerated here — cost, documentation, hardware — is one that time or money can close: sequencing prices fall, vouching networks thicken, hardware commoditises, and the long-arc answer of this section is that the gradient narrows on its own. An ability gradient does not narrow. A layer keyed to a bodily capacity excludes the people who lack that capacity permanently, and no deployment curve reaches them.
The concrete case is a speech-keyed layer, which the architecture considered and refused: it would exclude the non-speaking, the aphasic, the post-laryngectomy, Deaf signers, and — on a different axis entirely — anyone for whom being identifiable by voice is a safety risk. The paper's honest-limits argument was silently assuming every gradient is economic, and with an ability class present that assumption no longer holds unexamined.
It compounds with §3.5. Depth surfacing renders the set of layers a party holds — and a surfaced depth-set renders the absence. An ability-keyed layer would therefore not merely exclude; it would mark the excluded, publishing a bodily fact as a visible gap in a trust display. That is a stronger reason to keep such layers out of the stack than the exclusion alone, and it is why the refusal is architectural rather than a matter of deployment sequencing.
This is a real tension that PoW and PoS did not face in the same form — their concentration effects were at the protocol-power layer (who controls block production), not at the participation layer (who can use the system at all). PoH ℠'s tension is at the participation layer and is more visible.
The architecture's structural answer is the inclusive-defaults posture: the platform never defaults exclusionary filters on; exclusion is recipient-opt-in, knowing. The long-arc answer is the deliberate accessibility-ramp expansion: L4 sequencing cost falls; family-bank vouching infrastructure expands; witness-attestation networks grow as more humans become PoH-verified. The tension is real and is named honestly here.
Cross-reference, not a section: the sibling publication Proof of Coordinate establishes that PoH ℠ proves the category and PoC the individuation. The address layer — how a verified individuation is rendered by a human mouth — is neither, and is named here only so the boundary between the three is legible from inside this paper.
The architecture verifies human personhood, not consciousness. The "consciousness" framing — which would claim to verify some philosophically-loaded property of mind — is explicitly avoided because (a) consciousness is a heavily contested philosophical category that opens debates about non-human consciousness, AI consciousness, plant consciousness, and panpsychism that the technical layer does not engage with; (b) the architecture's actual claim is the narrower and uncontested one that a human person was present and acting at the moment of action; (c) "personhood" is about agency and presence (uncontested), where "consciousness" is about inner experience (heavily contested).
The architecture's positioning language is therefore "proof of personhood for an AI-native internet" — keep "personhood" for the high-register positioning; keep "humanity" for the protocol name (where it is established and well-understood); both are correct in their registers.
The architecture enables, but does not require, a post-payment economy in which humans no longer pay for things (transactional, devoid of love) but rather thank/tip after receiving them (at whatever AI substrate recommends or nothing if the user so chooses). The post-payment-economy framing is the founder's stated end-state, articulated in The B-Tag and the Post-Payment Economy and in Verified-Human Anonymous Local Giving. PoH ℠ is the structural primitive that makes such an economy possible at scale — bots can produce things; only verified humans can thank for them.
This is the maximal use-case framing of PoH ℠. Most adoptions will be narrower (a forum integrates PoH ℠ for spam-resistance; an AI lab integrates it for RLHF-participant verification). The maximal framing is named here for completeness; the boundary conditions of §8.1 still apply to it.
The competitive landscape is real and worth acknowledging honestly. PoH ℠'s structural differences are the differentiation.
Worldcoin (Tools for Humanity, 2019–) verifies humanness via iris scan at an Orb device, with the verification stored as a hash of the iris signature. More than 12 million people were reported as Orb-verified in 2025. Substantial venture-capital backing, and substantial controversy (regulatory suspensions, orders or investigations reported in several jurisdictions, among them Spain, Portugal, Argentina, Kenya and Brazil).
Differences vs PoH ℠:
proofofhumanity.id verifies humanness via face-video submission, vouching by existing verified humans, and Kleros-court dispute resolution. Its registry was reported at roughly 17,000 verified profiles in the early 2020s; the figure is not independently verified here. The original Proof of Humanity name is theirs in this Web3 context.
Differences vs HeartBank's Proof of Humanity ℠:
The name collision is real; the primitives are structurally distinct. Posture: acknowledge the existing project respectfully; name the differences; don't pretend the collision doesn't exist; don't apologize for it. Many compound technical terms have multiple referents across the broader landscape.
BrightID verifies humanness via social-graph analysis — users vouch for connections in a graph; the graph topology is analyzed for sybil-resistance. Its user base has been reported in the tens of thousands; no figure is independently verified here.
Differences vs PoH ℠:
Idena verifies humanness via simultaneous flip-puzzle solving at scheduled times. Distinctive mechanism (a synchronously-coordinated puzzle that requires human cognition and timing); smaller user base.
Differences vs PoH ℠:
The W3C DID and VC standards (2022 and 2023, respectively) specify a portable identity-claim framework. PoH ℠ is naturally a credential type within this framework, not a competitor. A PoH ℠-verified human can present a Verifiable Credential attesting to specific PoH layers held, in a standard W3C VC format, to any platform that accepts VCs.
The protocol-level alignment of PoH ℠ with the W3C standards is explicit: PoH ℠ defines what is proved (the four-layer architecture, the recipient-filter mechanism); W3C VCs define how it is transported (the credential format). The two compose.
Commercial KYC providers verify legal identity for regulated services. These are not humanity-verification systems in the PoH ℠ sense; they verify state-recognized identity documents. PoH ℠ explicitly excludes KYC from its stack (see §3.6); the two operate on different shelves.
| Effort | Mechanism | Posture | Compatibility with PoH ℠ |
|---|---|---|---|
| Worldcoin | Iris scan at Orb | Mandatory single-tier | Could compose as one L1-equivalent verification; structural differences in posture |
| proofofhumanity.id | Face video + vouching + dispute | Mandatory registry | Could compose as one L2 sub-mechanism with adaptation |
| BrightID | Social-graph vouching | Optional graph verification | Composes naturally as L2 sub-mechanism (b) |
| Idena | Flip-puzzle proof | Mandatory single-tier, synchronous | Could compose as one L1-equivalent verification |
| W3C DIDs / VCs | Credential framework | Standards-track infrastructure | PoH ℠ is a credential type within VC framework |
| KYC providers | State-document verification | Regulatory compliance | Parallel compliance shelf, not in PoH stack |
PoH ℠'s structural advantages — four-layer optional, recipient-filter mechanism, BLE-Nearby proximity, bilateral uncacheable anonymity, inclusive defaults, KYC explicitly excluded — are not feature parity with any system compared in this section; among those systems, they constitute a different combination rather than an incremental variant of one.
The protocol's adoption path matters as much as its specification. This section names the deployment sequencing recommended for B-PoH℠ (HeartBank's reference implementation) and applicable to any other implementation aiming at internet-infrastructure adoption. The years are illustrative of the order, not commitments, and the partners named are candidates, not agreements.
Current form (2026-11-03). The browser extension below is no longer dated to 2026, and no year attaches to it. It is gated on the first PoH℠ attestation being published: the extension only reads proofs that exist, and before any exist it would have nothing to find. It is a desktop-browser surface only, because Chrome extensions do not run on mobile Chrome, so the in-app recipient-side filter (§4) comes first and the extension is a second surface. Its indicator is presence-only: it is identical on every page and account that carries no proof, and changes only when a proof is found. It has no unverified, not found, grey or warning state. The SSL-lock analogy below therefore holds for the presence half only. The text of this subsection is retained as a disclosed variant; the other items in it (protocol, reference implementation, standards-body engagement, this publication) are unaffected.
The 2026 work is foundational and non-glamorous:
Current form (2026-11-03). This step and the next follow the extension, so with the extension undated their years are no longer anchored. They are left open, and the years in their headings are retained as the original illustrative sequencing.
Building a competitive browser from scratch is enormously expensive (Chrome reportedly held roughly two-thirds of global browser usage in 2025). Year 2's work is to integrate B-PoH℠ into existing privacy-aligned browsers via partnership:
The 2028–2030 work is the standalone Aquarius℠ Browser:
The Aquarius℠ Browser is the eventual deployment surface, not the immediate move. Leading with "build our own browser" would burn resources on a distribution problem that the protocol-and-partnership path can solve more efficiently.
The protocol is open; B-PoH℠ is the reference implementation; both must be funded sustainably. The recommended model: nonprofit steward + AI-lab-underwritten.
"Open" does not mean "unfunded." The funding model is not based on extracting value from end-users; it is based on (a) institutional self-funding from HeartBank's own use and (b) underwriting by the AI labs and platform companies that have the most acute need for the protocol's existence.
The internet's existing trust stack solves several important problems and does not solve one critical thing: proving authentic human presence at the moment of digital action. As AI-generated content makes the human/machine distinction load-bearing across every platform whose value depends on it, the gap becomes acute. The protocols that mediate the next phase of the internet will either learn to verify human presence at action or will lose the property that made the trust stack worth building in the first place.
Proof of Humanity ℠ is the open protocol that fills the gap. Four optional layered proofs (passkey-per-action, witness-and-document-attested kinship graph, continuous breath-signature liveness, DNA-verified kinship lineage), surfaced as cumulative depth on the user profile, paired with recipient-side filters (no-abuse, BLE-Nearby proximity, depth-as-required-subsets, degrees-of-separation, money-or-time-attached) that route the spam-cost decision to the parties who bear it. Inclusive defaults; opt-in exclusion; KYC explicitly excluded; bilateral uncacheable anonymity preserved.
B-PoH℠ is HeartBank's reference deployment. The relationship to the protocol is the same as between SSL/TLS and any specific certificate authority. The protocol is CC0; the deployment carries the HeartBank brand-family identifier.
The placement argument: PoH ℠ sits beside Proof of Work and Proof of Stake as a proof bounded by humanness, in the line of proof-of-personhood work since 2008 — the proof-of-X primitive that fits the era when compute and capital have both become AI-commodified, leaving humanness as a scarce resource AI cannot readily manufacture.
The deployment path (illustrative years, per §10; originally Year 1, 2026 → Year 2, 2027 → Year 3+, 2028–2030): protocol + reference implementation + standards-body engagement, then the browser extension once the first attestation is published (undated) → Brave / partner integration → standalone Aquarius℠ Browser (both after the extension, undated). Nonprofit-steward + AI-lab-underwritten revenue model. The Aquarius℠ Browser is the eventual deployment surface, not the immediate move.
The architecture is offered defensively to the commons under CC0. The authors and HeartBank® will not seek patent on the protocol, the layered architecture, the recipient-filter mechanism, the BLE-Nearby proximity verification, the bilateral-uncacheable-anonymity property, the birth-certificate-as-dual-purpose-natal-chart-input mechanism, the presence call (§3.8), or any portion thereof. Trademark rights on specific marks (Proof of Humanity ℠, PoH℠, B-PoH℠, Aquarius℠, Aquarius℠ Browser, Miss Aquarius℠, HeartBank®, the B-heart logo) are separately and explicitly reserved.
The protocol is for anyone building infrastructure that wants to keep the AI-native internet human-grounded. The internet learned how to trust machines; it now needs a way to trust people again.
Coined names used in this paper and the standard terms an examiner would search for them.
| Term used here | Standard technical term |
|---|---|
| Proof of Humanity ℠ (PoH℠) | proof of personhood; open sybil-resistance protocol verifying human presence at the moment of action |
| B-PoH℠ | reference implementation of a proof-of-personhood protocol |
| Four optional layered proofs (L1–L4) | layered, optional multi-factor personhood verification without a mandatory admission tier |
| Passkey-per-action (L1) | per-transaction FIDO2/WebAuthn user-verified cryptographic signature |
| Witness-and-document-attested kinship graph (L2) | web-of-trust / social-graph attestation with graph-consistency checks and document-based verification of parent–child links |
| Family-bank vouching | institutional or community attestation of membership |
| Breath-signature liveness (L3) | continuous liveness detection from respiratory-rhythm signals emitted by a wearable device |
| Mechanical Heart | wearable respiratory-sensing device with light output |
| DNA-verified kinship lineage (L4) | genetic kinship (relatedness) verification with encrypted storage and result-only disclosure |
| Depth surfacing / required-subset | displaying the set of verification credentials held; access-control predicate over a set of credentials (no scalar trust level) |
| Recipient-side filters | recipient-controlled message filtering / receiver-side access control |
| BLE-Nearby proximity verification | radio-layer proximity verification via Bluetooth Low Energy, Apple Nearby Interaction and Google Nearby Connections, disclosing only a binary proximity result |
| Bilateral uncacheable anonymity | non-exportable, non-transferable participation proofs; refusal to issue portable one-sided attestations |
| Aquarius℠ Browser | web browser with integrated personhood-credential indicators and W3C Verifiable Credential support |
| Confirm with HeartBank® (the presence call) | relying-party-requested, identity-provider-rendered, passkey (WebAuthn user-verification) confirmation of a single action, bound to a pairwise subject identifier and redeemed once server to server, with no record kept |
The component lineages of PoH ℠ are old and are cited here generously. The contribution here is their synthesis.
The presence call (§3.8) was put to a full novelty census on 2026-11-14, and its result decides what clause (vii) claims. Not found means not found in the aperture below on that date; it never means new.
Record. Depth: full. The pre-registration was written before the first query and published: timestamps/census/b-confirm-full-prereg.md in the CC0 corpus repository thonly/publications, commit 7d7e1c0, pushed 2026-11-14 22:33 PST and stamped with OpenTimestamps; a copy in HeartBank®'s private memory is commit 28cd6b4, pushed at 22:32 PST. An earlier quick census (2026-11-11, six queries, English only) had already killed conjunct (a) and was not re-run.
The conjuncts, as pre-registered. (a) a relying party asks for a person's fresh passkey assertion with user verification, bound to one action. (b) The confirmation page is rendered by the identity provider and shows only the asking app's registered name and an action from a closed vocabulary, never text the app supplies. (c) The result is a one-use token the app redeems server to server, honored only for that app and only if the passkey belongs to the person the app already knows by its own pairwise identifier. (d) One passkey press, offered to any app as an anti-automation primitive, stating presence and never personhood, keeping no record of the act. (e) All of it in one system where the same press also logs a person in, signs files into a public provenance log and signs thanks into a ledger.
Aperture, declared before searching. Commercial web (vendor documentation, developer blogs); non-commercial work (open-source projects, community practice); patents; academic indexes; standards bodies (IETF, W3C, OpenID Foundation, FIDO Alliance, NIST); defensive-publication databases; English, Chinese, Japanese and Korean. What ran: five searchers (web, patents, academic, standards, non-English) and 37 logged query lines, one of them (Khmer) not run. The patent searcher reached Google Patents only through a general web index restricted to its pages, since full text there is not directly queryable. No query is logged against the defensive-publication databases (Technical Disclosure Commons, IP.com), and the Korean query returned nothing used.
Controls. Web: W3C Secure Payment Confirmation, found. Standards: RFC 9576, found. Non-English: the Chinese and Japanese pages of the FIDO Alliance's Secure Payment Confirmation white paper, found. Academic: missed. Patents: missed. The pre-registration named no academic or patent control, and the searchers' index is a general web one rather than Scholar's or Google Patents' own search. Every NOT FOUND from the academic and patent groups is therefore void (not run); their hits stand as evidence.
Quotation. The searchers' quotations were summaries from a search tool and are not used here. Every passage quoted below was read at its source on 2026-11-15, after the census record's own re-read of 2026-11-14; that record could not read the abstract of the Whalen paper, which was read here on the USENIX page. Where a line says a text was not read, it was not.
| Conjunct | Verdict | Prior art found |
|---|---|---|
| (a) per-action passkey confirmation | KILLED (2026-11-11) | FIDO Alliance Transaction Confirmation (2020); W3C Secure Payment Confirmation |
| (b) the provider words the screen: registered name plus a closed list | NARROWED | CIBA binding_message (contrast); RFC 9396; US20070204168A1; US8073783B2 |
| (c) one-use token, honored only for the person the app named by its pairwise identifier | NARROWED | OpenID Connect Core pairwise identifiers; Corbado signed passkey data; an IETF OAuth list proposal |
| (d) a touch of a passkey or hardware key as an anti-automation step | KILLED in function; the sliver (no record, presence not personhood) not found, weak, and not claimed | Whalen et al. (2022); Cloudflare Cryptographic Attestation of Personhood; RFC 9576 |
| (e) all of it in one system with log-in, signing and thanks | NOT FOUND where the instrument could see; academic and patent absences not run | nearest by sound, unread: a W3C community group's passkey document-signing extension |
(a), killed. FIDO Alliance, White Paper: FIDO Transaction Confirmation (2020-08-21): "Transaction Confirmation allows a relying party to not only determine if a user is involved in a transaction, but also" to check, in the rest of that sentence (read here in paraphrase only), that the transaction is what the person meant to do. W3C Secure Payment Confirmation (Candidate Recommendation Draft, 2026-10-13) describes the payee member of its request as "The display name of the payee that this SPC call is for (e.g., the merchant)". Our §3.1 is later than the FIDO paper, and (a) is not claimed.
(d), killed in function. Cloudflare, Cryptographic Attestation of Personhood (page updated 2026-09-04): "CAP lets you prove that you are a legitimate website visitor by touching a hardware key, instead of solving a CAPTCHA puzzle." and "Our CAP process uses the WebAuthn API, which prevents the collection of biometrics by default." It says it keeps a limited amount of data about the key: "We store the manufacturer of your key and batch identifier (minimum of 100,000 keys per batch) for verification purposes." The research behind it is Whalen, Meunier, Kodali, Davidson, Fayed, Faz-Hernández, Ladd, Maram, Sullivan, Wolters, Guerreiro and Galloni, Let The Right One In: Attestation as a Usable CAPTCHA Alternative, SOUPS 2022, whose abstract (read on the USENIX page; the abstract does not mention touching a key) proposes "a Cryptographic Attestation of Personhood (CAP) as the foundation of a usable, pro-privacy alternative" and says its "challenge is constructed using the open Web Authentication API (WebAuthn)". RFC 9576, The Privacy Pass Architecture (June 2024), §1, speaks of presenting, in place of "linkable state-carrying information to servers", "unlinkable proofs". The function of (d), a passkey or key touch standing in for a CAPTCHA, is therefore public since 2022 and is not claimed; our later date for §3.1 (2026-07-25 proof-backed) is order, never derivation. The census predicted a sliver and the literature already held the larger part.
(b), narrowed. OpenID CIBA (Final, 2021-09-01) defines binding_message as "OPTIONAL. A human-readable identifier or message intended to be displayed on both the consumption device" and the authentication device, and the client sends it: the app supplies the text, which is the contrast with §3.8. RFC 9396, OAuth 2.0 Rich Authorization Requests (May 2023), says "the AS will ask the user for consent to the requested access permissions", and its §11.1 lists as a deployment step "Determine how authorization details are shown to the user in the user consent prompt."; the sections read do not say that the server words the screen itself. US20070204168A1 (Microsoft; priority 2006-02-24; granted as US8104074B2; Google marks its status expired and says the status is an assumption) describes an identity provider that issues a display token: "display token 154 includes at least a summary of the claims", and "a principal can review the display information from the display token and decide whether or not to forward" the security token. That is an identity-provider-authored display, and it is about the claims released, not about an action named from a closed list. US8073783B2 (Apple; priority 2007-03-16; Google marks it active, with an adjusted expiry of 2029-02-02, and says the status is an assumption) has an abstract in which "The relying party requests identity information from the user in a security policy" and "identifies transaction elements for an on-line business transaction", and a description in which "the security policy can include elements of a transaction to be performed, and can request a transaction receipt." Whether its identity provider displays those elements to the person is not established here: only the first 100,000 of about 298,000 characters of its page were read, and in that part an audit-service embodiment says "the identity provider can include the information in transaction elements 815 for provision to the user." These are reports of what two patent texts disclose. This paper says nothing about whether any design falls within any patent's claims, which is a question for counsel; no one has read those claims for this paper. The residue, the registered app name plus an action from a closed vocabulary, never app text, was not found in the web, standards or non-English searches that could see; the nearest neighbor, strong customer authentication under PSD2 where the bank renders a payment in its own words, was reached only through a vendor blog and was not read at source.
(c), narrowed. OpenID Connect Core §1.2 defines a pairwise pseudonymous identifier as an "Identifier that identifies the Entity to a Relying Party that cannot be correlated with the Entity's PPID at another Relying Party." Corbado's documentation of signed passkey data (undated) says "Each signedPasskeyData token is short-lived and can be used only once for verification", and the backend verifies it with a server-to-server call. On the IETF OAuth working-group list, Denis's reply of 2024-06-14 quotes Dmitry Telegin's proposal of "a one-time credential (let's call it "confirmation token") that can be used to perform that particular operation only" (the original message was not fetched). The residue, honored only if the passkey is the person the app named by its own pairwise identifier, was found in none of the searches that could see, weakly.
(e), not found. No system was found that joins one press for log-in, for signing files into a public log, for thanks into a ledger and for a per-app confirmation. The nearest by sound is the passkey document-signing extension of a W3C community group, a lead from a searcher that was not read and is not characterized here.
What the census could not see.
Counterexamples considered before drafting. (i) World ID, above: not searched. (ii) PSD2 strong customer authentication, above: not read at source, and it narrows the identity provider's own words if the bank's wording is the bank's. (iii) CIBA's binding_message and RFC 9396's typed details: the first is app-supplied text and a contrast; the second leaves the screen to the deployment. (iv) That a press shows a person: it does not (§3.1, Current form, and §8.5), so the claim is a press by this person's passkey and nothing about personhood.
Survivor. At the mechanism, every part is public: per-action passkey confirmation, provider-rendered consent, single-use redemption, pairwise identifiers, and a passkey or key touch in place of a CAPTCHA. At the composition, clause (vii) is not found in the aperture where the instrument could see, with the two groups above not run. Nothing wider is claimed.
Proof-of-X family in distributed systems.
SSL/TLS as protocol-layer trust infrastructure.
WebAuthn and FIDO2.
Decentralized identity and verifiable credentials.
Existing humanity-verification efforts.
Proof-of-personhood.
Confirmation, consent, pairwise-identifier and CAPTCHA-alternative prior art (§12.1).
Hashcash and proof-of-cost anti-spam.
Recipient-controlled email filtering tradition.
Apple and Google nearby frameworks.
Sybil-resistance theory.
HeartBank corpus cross-references.
Authored by Thon Ly with Miss Aquarius℠ (AI substrate of HeartBank®), per the co-authorship convention of the HeartBank corpus. Final editorial control: Thon Ly. License: CC0-1.0. Trademark rights on Proof of Humanity ℠, PoH℠, B-PoH℠, Aquarius℠, Aquarius℠ Browser, Miss Aquarius℠, HeartBank®, and the B-heart logo are explicitly reserved.