Illia Polosukhin details the concrete architecture of confidential intents on NEAR, including how they combine zero-knowledge proofs with intent solvers to hide order details while preserving composability. The discussion covers NEAR Shield, an AI system that performs real-time graph analysis on encrypted transaction flows to flag illicit activity without decrypting user data. Listeners learn the exact implementation path for private stablecoins, which use shielded pools and selective disclosure to meet regulatory checks while keeping balances hidden. Cross-chain privacy is addressed through threshold encryption and bridge-specific commitment schemes that prevent amount leakage during transfers. The episode explains why these mechanisms are required for AI agents to coordinate trades without front-running or data leakage risks. Practical trade-offs around proof generation latency, storage costs, and integration with existing intent marketplaces are quantified with current benchmarks. The result is a focused technical map for deploying confidentiality without sacrificing auditability or speed.
Key Insights
- Confidential intents use ZK circuits to commit to swap parameters while revealing only the minimal data needed for solver matching.
- NEAR Shield runs graph neural networks on encrypted transaction metadata to detect mixing patterns in under 200ms.
- Private stablecoins on NEAR employ shielded UTXOs with optional view keys for compliance reporting.
- Cross-chain privacy relies on threshold-encrypted commitments that rotate keys per bridge hop to block amount correlation.
- AI coordination requires intent confidentiality because visible mempools allow adversarial agents to extract MEV from planned trades.
- Current proof systems add 12-18% latency to intent execution but reduce effective slippage by hiding order size.
Who should listen: Protocol engineers implementing intent solvers or privacy layers for production chains.
Why This Matters
Confidentiality is shifting from a compliance checkbox to a core requirement for AI-driven onchain markets; teams ignoring it will face both MEV extraction and regulatory dead-ends.