The Two Families

zkSNARKs (Succinct Non-interactive Arguments of Knowledge):

• •Small proof size (~200 bytes)
• •Fast verification (~10ms)
• •Requires trusted setup ceremony
• •Used by: Zcash, Tornado Cash, zkSync

zkSTARKs (Scalable Transparent Arguments of Knowledge):

• •Larger proof size (~50KB)
• •Slower verification (~100ms)
• •No trusted setup required (transparent)
• •Used by: StarkNet, Polygon Miden
• •Quantum-resistant (future-proof)

When to Use Which

Application 1: Blockchain Scaling (ZK Rollups)

How ZK Rollups Work

1. •Thousands of transactions execute off-chain
2. •A ZK proof attests that all transactions are valid
3. •Only the proof + compressed state diff posted to L1
4. •L1 verifies the proof in ~200K gas (vs millions for re-execution)

Result: 100-2000x throughput improvement while inheriting L1 security.

Major ZK Rollups

• •zkSync Era: $4.2B TVL, EVM-compatible, native account abstraction
• •Starknet: $2.1B TVL, Cairo language, highest computational throughput
• •Scroll: $1.8B TVL, bytecode-level EVM equivalence
• •Linea: $1.5B TVL, ConsenSys-backed, type 2 zkEVM
• •Polygon zkEVM: $1.2B TVL, equivalence-focused

Application 2: Identity and Compliance (ZK-KYC)

The Privacy-Compliance Paradox

Regulators demand identity verification. Users demand privacy. ZK proofs solve both:

Without ZK: Share passport, address, bank statements → company stores everything → data breach risk

With ZK: Prove "I am over 18 AND not on sanctions list AND resident of allowed country" → verifier learns nothing else → zero data stored

Real Implementations

• •Polygon ID: ZK credential verification with selective disclosure
• •Worldcoin: ZK proof of unique personhood (no biometric data shared)
• •zkPass: ZK-verified web credentials (prove social media ownership without API access)
• •Sismo: ZK attestations from on-chain and off-chain data

Application 3: Verifiable AI/ML

The AI Trust Problem

AI models are black boxes. How do you trust that:

• •A model was trained on the claimed dataset?
• •Inference used the actual model (not a cheaper substitute)?
• •Output wasn't tampered with post-generation?

ZK-ML: Verifiable Machine Learning

ZK proofs can verify ML inference without revealing model weights:

1. •Model runs inference and generates output
2. •ZK proof attests that output came from the specific model with specific input
3. •Verifier confirms without seeing model parameters

Applications:

• •AI oracle verification: Prove an AI prediction came from a specific model
• •Regulatory compliance: Prove model doesn't use prohibited features (age, race)
• •Audit trail: Cryptographic proof of which model generated which decision
• •Competitive markets: Prove model performance without revealing architecture

Current Limitations

• •Proving large neural networks is computationally expensive
• •Limited to smaller models (currently ~10M parameters practically)
• •Active research from EZKL, Modulus Labs, and academic institutions

Application 4: Supply Chain Verification

Beyond Traditional Tracking

ZK proofs in supply chain go beyond "where is my package":

• •Prove origin without revealing supplier: Company proves products meet sourcing requirements without exposing trade secrets
• •Verify certifications: Prove organic/fair-trade certification without revealing full audit
• •Cross-border compliance: Prove goods meet import requirements without sharing commercial details

Application 5: Voting and Governance

Private, Verifiable Voting

ZK-based voting enables:

• •Vote privacy: No one can see how you voted (prevents coercion and vote buying)
• •Verification: Anyone can verify the total count is correct
• •Eligibility: Prove you're eligible to vote without revealing identity
• •Coercion resistance: Even under pressure, you can't prove how you voted

Implementations: MACI (Minimal Anti-Collusion Infrastructure) used in Gitcoin grants and DAO governance.

Getting Started with ZK Development

Developer Tools

Learning Path

1. •Conceptual: Understand polynomial commitments and arithmetic circuits
2. •Practical: Write simple circuits in Circom or Noir
3. •Application: Build a ZK-verified credential or voting system
4. •Advanced: ZK-ML, custom proof systems, recursive proofs

Key Takeaways

1. •ZK proofs power $10B+ TVL in rollups — they're production infrastructure, not just theory
2. •ZK-KYC solves the privacy-compliance paradox — prove regulatory compliance without exposing personal data
3. •Verifiable AI (ZK-ML) is the next frontier — prove model integrity without revealing proprietary weights
4. •zkSTARKs are quantum-resistant — future-proof your ZK infrastructure against quantum computing threats

FAQ

Do I need to understand the math to use ZK proofs?

No. Modern tools like Circom, Noir, and Polygon ID abstract the cryptography. You define what you want to prove in a high-level language, and the tooling handles proof generation and verification. Understanding the concepts helps but isn't required for application development.

How expensive is ZK proof generation?

It depends on circuit complexity. Simple proofs (age verification) generate in milliseconds. Complex proofs (ZK rollup batches) take seconds to minutes and require significant compute. Costs are dropping rapidly — proof generation is 10x cheaper than 2 years ago.

Are zero-knowledge proofs truly zero-knowledge?

In the cryptographic sense, yes — the verifier learns nothing beyond the truth of the statement. In practice, metadata (timing, frequency, gas usage) can sometimes leak information. Good system design minimizes these side channels.

Find ZK development teams on The Signal directory.

The Two Families

zkSNARKs (Succinct Non-interactive Arguments of Knowledge):

• •Small proof size (~200 bytes)
• •Fast verification (~10ms)
• •Requires trusted setup ceremony
• •Used by: Zcash, Tornado Cash, zkSync

zkSTARKs (Scalable Transparent Arguments of Knowledge):

• •Larger proof size (~50KB)
• •Slower verification (~100ms)
• •No trusted setup required (transparent)
• •Used by: StarkNet, Polygon Miden
• •Quantum-resistant (future-proof)

When to Use Which

Application 1: Blockchain Scaling (ZK Rollups)

How ZK Rollups Work

1. •Thousands of transactions execute off-chain
2. •A ZK proof attests that all transactions are valid
3. •Only the proof + compressed state diff posted to L1
4. •L1 verifies the proof in ~200K gas (vs millions for re-execution)

Result: 100-2000x throughput improvement while inheriting L1 security.

Major ZK Rollups

• •zkSync Era: $4.2B TVL, EVM-compatible, native account abstraction
• •Starknet: $2.1B TVL, Cairo language, highest computational throughput
• •Scroll: $1.8B TVL, bytecode-level EVM equivalence
• •Linea: $1.5B TVL, ConsenSys-backed, type 2 zkEVM
• •Polygon zkEVM: $1.2B TVL, equivalence-focused

Application 2: Identity and Compliance (ZK-KYC)

The Privacy-Compliance Paradox

Regulators demand identity verification. Users demand privacy. ZK proofs solve both:

Without ZK: Share passport, address, bank statements → company stores everything → data breach risk

With ZK: Prove "I am over 18 AND not on sanctions list AND resident of allowed country" → verifier learns nothing else → zero data stored

Real Implementations

• •Polygon ID: ZK credential verification with selective disclosure
• •Worldcoin: ZK proof of unique personhood (no biometric data shared)
• •zkPass: ZK-verified web credentials (prove social media ownership without API access)
• •Sismo: ZK attestations from on-chain and off-chain data

Application 3: Verifiable AI/ML

The AI Trust Problem

AI models are black boxes. How do you trust that:

• •A model was trained on the claimed dataset?
• •Inference used the actual model (not a cheaper substitute)?
• •Output wasn't tampered with post-generation?

ZK-ML: Verifiable Machine Learning

ZK proofs can verify ML inference without revealing model weights:

1. •Model runs inference and generates output
2. •ZK proof attests that output came from the specific model with specific input
3. •Verifier confirms without seeing model parameters

Applications:

• •AI oracle verification: Prove an AI prediction came from a specific model
• •Regulatory compliance: Prove model doesn't use prohibited features (age, race)
• •Audit trail: Cryptographic proof of which model generated which decision
• •Competitive markets: Prove model performance without revealing architecture

Current Limitations

• •Proving large neural networks is computationally expensive
• •Limited to smaller models (currently ~10M parameters practically)
• •Active research from EZKL, Modulus Labs, and academic institutions

Application 4: Supply Chain Verification

Beyond Traditional Tracking

ZK proofs in supply chain go beyond "where is my package":

• •Prove origin without revealing supplier: Company proves products meet sourcing requirements without exposing trade secrets
• •Verify certifications: Prove organic/fair-trade certification without revealing full audit
• •Cross-border compliance: Prove goods meet import requirements without sharing commercial details

Application 5: Voting and Governance

Private, Verifiable Voting

ZK-based voting enables:

• •Vote privacy: No one can see how you voted (prevents coercion and vote buying)
• •Verification: Anyone can verify the total count is correct
• •Eligibility: Prove you're eligible to vote without revealing identity
• •Coercion resistance: Even under pressure, you can't prove how you voted

Implementations: MACI (Minimal Anti-Collusion Infrastructure) used in Gitcoin grants and DAO governance.

Getting Started with ZK Development

Developer Tools

Learning Path

1. •Conceptual: Understand polynomial commitments and arithmetic circuits
2. •Practical: Write simple circuits in Circom or Noir
3. •Application: Build a ZK-verified credential or voting system
4. •Advanced: ZK-ML, custom proof systems, recursive proofs

Key Takeaways

1. •ZK proofs power $10B+ TVL in rollups — they're production infrastructure, not just theory
2. •ZK-KYC solves the privacy-compliance paradox — prove regulatory compliance without exposing personal data
3. •Verifiable AI (ZK-ML) is the next frontier — prove model integrity without revealing proprietary weights
4. •zkSTARKs are quantum-resistant — future-proof your ZK infrastructure against quantum computing threats

FAQ

Do I need to understand the math to use ZK proofs?

No. Modern tools like Circom, Noir, and Polygon ID abstract the cryptography. You define what you want to prove in a high-level language, and the tooling handles proof generation and verification. Understanding the concepts helps but isn't required for application development.

How expensive is ZK proof generation?

It depends on circuit complexity. Simple proofs (age verification) generate in milliseconds. Complex proofs (ZK rollup batches) take seconds to minutes and require significant compute. Costs are dropping rapidly — proof generation is 10x cheaper than 2 years ago.

Are zero-knowledge proofs truly zero-knowledge?

In the cryptographic sense, yes — the verifier learns nothing beyond the truth of the statement. In practice, metadata (timing, frequency, gas usage) can sometimes leak information. Good system design minimizes these side channels.

Find ZK development teams on The Signal directory.