Cary Greenwood
Back to blog
•
10 min read
Human Verification + Human-Created Media Authentication (Crypto + Provenance)

Human Verification + Human-Created Media Authentication (Crypto + Provenance)

Executive summary

You’re asking two related questions:

  1. How to verify a user is human (not an AI/bot) across web/image/video workflows.
  2. How to verify an asset was human-created or authentically captured, similar to supply-chain provenance in crypto.

Short answer: there is no single perfect mechanism today. The strongest practical design is a stacked trust model:

  • Proof-of-personhood / anti-Sybil layer (who is interacting)
  • Capture/provenance layer (how/where media was created/edited)
  • Cryptographic attestation + public audit trail (tamper evidence)

Existing solution landscape

A) Bot/human verification on websites (low friction)

1) Cloudflare Turnstile

  • Invisible/low-friction CAPTCHA alternative.
  • Uses browser/device signals and challenge-response, designed to reduce explicit CAPTCHA puzzles.
  • Good for signup/login/form abuse defense.
  • Limitation: proves “likely human interaction right now,” not durable personhood identity.

Best for: consumer web apps that need anti-bot at scale with low UX friction.

B) Proof-of-personhood (PoP) / anti-Sybil in crypto

2) World ID / Orb (World)

  • Biometric uniqueness verification (face/iris capture flow) for one-person uniqueness claims.
  • Designed to issue a reusable proof of unique human.
  • Strong anti-Sybil potential, but high social/regulatory sensitivity around biometrics and governance trust assumptions.

3) Proof of Humanity (Kleros)

  • Registry model using profile + photo/video + social vouching + challenge/arbitration.
  • Explicitly positioned for governance/airdrop anti-Sybil use cases.
  • More community-centric and governance-heavy than device-native methods.

4) BrightID

  • Social-graph based uniqueness (privacy-first claim, no traditional KYC dependency).
  • Better for communities with existing trust graph.
  • Weaker for cold-start environments without social connections.

5) Human Passport (formerly Gitcoin Passport)

  • Aggregates multiple “stamps”/signals to compute personhood/sybil-resistance scores.
  • Widely used in grants/airdrops/governance gating.
  • Strong composability; score quality depends on stamp quality and attack adaptation.

6) Humanity Protocol

  • Palm biometrics + verifiable credentials + ZK claims framing.
  • Ambition: reusable identity proofs with selective disclosure.
  • Emerging ecosystem; adoption and independent validation maturity should be assessed case by case.

C) Human identity verification from media (selfie/video/liveness)

7) Biometric liveness + selfie-to-ID vendors (Persona, iProov, FaceTec, etc.)

  • Typical flow: government ID + selfie/video + liveness/injection-attack checks.
  • Stronger than CAPTCHA for “is this a live person linked to an ID document?”
  • Limitation: privacy/compliance burden and ongoing deepfake race (must continuously update anti-spoofing).

Best for: KYC, regulated onboarding, high-value account actions.

D) Authenticating media origin/edit history (provenance)

8) C2PA / Content Credentials (industry standard)

  • Open spec for signed provenance metadata (origin, edits, tool assertions, AI-use disclosures).
  • Backed by broad ecosystem (Adobe, Microsoft, BBC, Google, OpenAI, etc.).
  • Think of it as a cryptographic nutrition label for media.
  • Limitation: not universal adoption yet; metadata can be stripped in some pipelines unless robustly embedded and checked.

9) Project Origin / IPTC Verified News Publisher

  • News/media workflows for publisher identity verification and tamper-evident provenance chains.
  • Focused on trust in journalism supply chains.

10) Truepic controlled capture + C2PA signing

  • “Secure capture” model: trusted app pipeline signs at capture time.
  • Good pattern for high-integrity evidence workflows (insurance, field operations, inspections).

E) Blockchain timestamping / immutable anchoring

11) OpenTimestamps (Bitcoin anchoring)

  • Proves a file hash existed before a time.
  • Useful to anchor manifests/claims without publishing raw media.
  • Limitation: timestamping alone does not prove the media is true or human-created; only that a hash existed.

12) Blockchain-native provenance projects (e.g., Numbers Protocol)

  • Attempt to combine media provenance with on-chain registration/discovery.
  • Useful in creator/NFT ecosystems; maturity and interoperability vary.

Can “supply-chain provenance” be applied to human media creation?

Yes — and this is likely the right conceptual model.

Map supply-chain concepts to media:

  • Raw material origin → original capture event (device, time, place, signer)
  • Transformation events → edit operations, tools, model usage declarations
  • Custody handoffs → who handled/published/licensed the asset
  • Final product verification → viewer-side verification of full chain integrity

The closest existing standard to this is C2PA, and crypto can augment it with public timestamping, DID/VC identity attestations, and auditable event logs.

Layer 1 — Human presence / personhood

Pick by risk level:

  • Low risk: Turnstile-style anti-bot + behavioral risk scoring.
  • Medium risk: wallet/personhood scoring (Human Passport/BrightID/PoH).
  • High risk: liveness + IDV + optional biometric uniqueness proof.

Layer 2 — Capture authenticity

  • Use trusted capture SDK or secure app pipeline.
  • Sign capture metadata at source (device key / org key).
  • Emit C2PA manifest at creation time.

Layer 3 — Edit provenance

  • Every transformation (crop, color, compositing, gen-AI insertion) appends signed assertions.
  • Preserve chain-of-custody events in C2PA-compatible history.

Layer 4 — Public audit anchoring

  • Anchor manifest hashes periodically to a public chain (BTC/ETH/L2) via timestamping/notarization.
  • Store only hashes + pointers, not sensitive media.

Layer 5 — Identity + selective disclosure

  • Use DID/VC claims for creator assertions (e.g., “verified journalist”, “verified contractor”).
  • Use ZK proofs where possible for privacy-preserving claim checks.

Layer 6 — Verification UX

  • Show trust state to end users:
    • Verified origin
    • Edit history complete/incomplete
    • Human capture confidence
    • Identity confidence level
  • Make “unknown/insufficient provenance” explicit rather than binary yes/no.

Proposed scoring model for “human-created asset confidence”

Use a weighted score (0–100) instead of binary truth:

  • Capture integrity (signed at source, secure pipeline)
  • Provenance continuity (no broken chain)
  • Identity assurance (none / social / KYC / biometric uniqueness)
  • Tamper evidence status
  • Publication attestation (known org/publisher cert)
  • Cross-source corroboration (optional)

This mirrors fraud/risk systems and degrades gracefully when data is partial.

Key design principles / pitfalls

Principles

  1. Prove claims, not identity by default (privacy-first).
  2. Use open standards first (C2PA, W3C credentials, WebAuthn where relevant).
  3. Separate personhood from content truth (a real human can still publish false content).
  4. Design for adversarial adaptation (deepfake and spoof techniques evolve fast).
  5. Support partial trust (confidence levels beat brittle binary decisions).

Pitfalls

  • Assuming blockchain timestamp = truth.
  • Over-centralizing trust in one issuer/oracle.
  • Ignoring metadata stripping during social reposting.
  • Shipping without verifier UX (users won’t inspect raw manifests).
  • Treating passkeys/auth as bot-proofing (different problem).

Suggested MVPs

MVP 1: “Human Capture Receipt” for creators

  • Mobile/web capture app
  • Signed C2PA manifest
  • Hash anchor on chain
  • Public verifier page

MVP 2: “Trusted Submission Gateway” for marketplaces/newsrooms

  • Human verification at upload (tiered)
  • Automated provenance policy checks
  • Risk score + reviewer queue

MVP 3: “Enterprise evidence mode”

  • Controlled capture SDK in field apps
  • Immutable chain-of-custody and role attestations
  • Audit exports for legal/compliance workflows

My recommendation for your direction

If your goal is strongest practical trust with crypto-native composability:

  1. Base protocol: C2PA manifest chain for media provenance.
  2. Identity layer: VC-based personhood attestations (with optional PoP provider integrations).
  3. Anchor layer: periodic on-chain hash commitments (timestamp + anti-tamper public audit).
  4. Policy engine: configurable confidence scoring and risk thresholds by use case.

This gives you supply-chain-like provenance for digital media while keeping room for privacy and multiple identity providers.

Transcript Addendum (Latest Google Doc, 2026-04-24)

The latest transcript (“Thoughts on Human Creativity in an increasingly AI World”) reinforces the core premise of this report:

  • In an AI-saturated future, human creative output may become scarce.
  • Society may increasingly view human creation as high-cost, high-intent, and risky.
  • Therefore, creators will need explicit systems for human-origin authentication and human creative preservation.

Why this matters to architecture

This transcript strengthens the need for two explicit design goals:

  1. Human-origin evidence (who created this, and with what human involvement).
  2. Human-intent continuity (how human choices/edits persisted through AI-assisted workflows).

Verbatim excerpt used as design input

Speaker 2 (01:33): “…human creative product creation, authentication, and verification to ensure that creative expression humans will not be lost through the creation of AI methodologies… I believe there should be a way to preserve and validate human creative potential.”

Design implication update

In addition to C2PA + on-chain anchoring, include an explicit Human Creative Intent Assertion in provenance manifests:

  • human_origin_claim: declared by creator identity/credential
  • human_edit_ratio: optional disclosed estimate of human vs AI transformation stages
  • human_final_approval: cryptographic sign-off by the human publisher before release

This keeps the system from only proving “untampered media” and moves it toward proving “meaningful human creative contribution.”

Consolidated conclusions from follow-up discussion

1) Blockchain alone does not solve “human-only”

Blockchain gives:

  • Immutable timestamping
  • Tamper-evident history
  • Public auditability

Blockchain does not inherently prove:

  • A human (vs AI/automation) created the asset
  • A human (vs bot/agent/script) created the on-chain record event

2) What must be added to enforce “human-only” claims

To move from provenance to human-authorship guarantees, the stack needs pre-chain enforcement:

  1. Human credential layer (PoP/KYC/VC depending on privacy target)
  2. Liveness-at-sign-time (not just one-time onboarding)
  3. Human-held secure keys (passkey/secure enclave/hardware signer)
  4. Policy-gated chain writes (on-chain write allowed only after attestation bundle validates)
  5. Optional co-attestation/witnesses for high-value media
  6. Challenge-and-penalty process (disputes, slashing/reputation loss for false human claims)

3) Does an end-to-end protocol already exist?

Conclusion: It exists in composable parts, but no broadly adopted single protocol was found that enforces both:

  • human-only creation, and
  • human-only blockchain record creation

4) What exists today (closest components)

  • Proof-of-human systems: World ID, Proof of Humanity, Human Passport, BrightID (different trust/privacy models)
  • Tamper-evident provenance systems: C2PA/Content Credentials, Project Origin patterns, OpenTimestamps, Numbers Protocol, Vbrick Verified Authentic
  • Identity assertions in provenance: CAWG identity assertions can bind credentials to content, but actor models are not strictly human-only by default

5) Practical product thesis (updated)

The strongest near-term product is a hybrid attestation pipeline:

  • C2PA for media lifecycle provenance
  • Personhood/liveness attestations for human presence
  • On-chain hash anchoring for public audit
  • Explicit human-intent assertions (human_origin_claim, human_edit_ratio, human_final_approval)

This is the most realistic path to “human-authenticated creativity” without waiting for a future monolithic standard.

Open questions to tighten next iteration

  1. Is your target use case consumer social, creator economy, enterprise evidence, or governance/airdrop anti-Sybil?
  2. Do you need anonymous uniqueness or real-world legal identity?
  3. Are you optimizing for minimal friction, maximal assurance, or regulatory compliance?
  4. Do you want chain-neutral anchoring (BTC + ETH), or specific ecosystem alignment?
  5. Should “human-only” be a hard gate, or a confidence score threshold with escalation review?