Trust Center

Verify what is required. Expose only what is necessary. Authorize every action. Record the decision, not the sensitive data. NEUS does not add compliance. It proves the architecture is operating according to its trust model.

​

Architecture

​

Zero-trust execution

Every action is authorized against identity, proof, policy, and delegated authority. Identity, ownership, login, or possession never automatically implies authority. Authority is checked against current policy and delegation at execution time.

​

Private by default

Proofs and receipts remain private unless explicitly shared. Public exposure, public indexing, IPFS, and chain anchoring are explicit choices. Off-chain is the default.

​

Minimum disclosure

A check returns the minimum claim or result required for the decision. The proof proves the decision without reproducing the underlying evidence. Apps and agents receive the result they need, not the sensitive source material.

​

Server-side credentials

Agents use connected tools without receiving the user’s secrets. Credentials are resolved and injected inside trusted server execution. They are never intentionally handed to the model, exposed in a tool schema, proof, browser response, or public record.

​

Auditable actions

Critical decisions produce proofs that can be inspected, verified, and revoked without publicly exposing the underlying data. Store enough to reconstruct who authorized what, under which policy, against which proof, and what outcome occurred.

​

Optional anchoring

Off-chain is the default. Public or blockchain publication requires explicit intent.

​

Invariants

These are NEUS invariants. Deviation from any of them is a bug.

• No implicit authority. Identity, ownership, login, or possession never automatically implies authority to perform an action. Authority is checked against current policy and delegation at execution time.
• Minimum disclosure. A check returns the minimum claim or result required for the decision. The proof proves the decision without unnecessarily reproducing the underlying evidence.
• Private and off-chain by default. Public exposure, public indexing, IPFS, and chain anchoring are explicit choices.
• Secrets never become agent context. Credentials are resolved and injected inside trusted server execution. They are never intentionally handed to the LLM, exposed in a tool schema, proof, browser response, or public proof.
• Evidence is not public proof. Private evidence can exist when it is necessary to make a proof auditable, but public consumers get the claim, status, or proof rather than the sensitive source material.
• Every privileged action has a policy decision. Subject plus resource plus action plus policy plus context produces allow, deny, or approval. The proof records the decision and relevant non-sensitive metadata.
• Delegation is narrower than ownership. Agents operate only inside explicitly delegated authority. Spend, secrets, tool access, and high-risk actions remain bounded.
• Auditability without surveillance. Store enough to reconstruct who authorized what, under which policy, against which proof, and what outcome occurred. Do not interpret “audit log” as permission to retain arbitrary request bodies, prompts, OAuth payloads, or provider responses.

NEUS provides evidence of control, not evidence about the person. That is stronger than calling it “privacy-first.”

​

A distinction that matters

Zero trust is not zero data. You sometimes legitimately need private state, encrypted secrets, session state, private evidence, or proofs to provide the product. The standard is purpose-bound, least-privilege, minimum necessary, protected, auditable, and deletable where applicable. Not “the server may never possess data.” For blockchain specifically, the EDPB’s final July 2026 guidance makes the off-chain-first and optional-anchor direction especially valuable.

​

One data model

Every feature (OAuth, Gmail, Drive, GitHub, payments, identity, ZK proofs, agents, MCP, API keys, AI inference, social connections, jobs) fits into one existing model. Evidence → Proof → Policy → Authority → Action → Receipt Not a Google system, an AI safety system, a wallet system, a GDPR system, an agent approval system, and an audit system. That fragmentation would destroy the advantage already built. The complete flow:

1. Sensitive evidence is examined inside the appropriate trusted boundary.
2. It is minimized into a claim or proof.
3. It is checked against policy and delegated authority.
4. The action is allowed, denied, or approved.
5. The proof records what was decided.
6. Sensitive evidence remains protected or disappears according to its retention requirement.

​

Privacy controls

Use this section only to choose visibility. Most builders should stay private.

​

Boundaries

• Proof issuance runs in NEUS.
• Return only what your product needs from checks.
• Private reads need the owner signed in (or your sharing rules).

​

Modes

Saved proof IDs follow the visibility you chose. privacyLevel decides private vs public. publicDisplay only matters for public proofs. Unlisted public still works for anyone with the id.

​

Defaults

For browser checkout, publish a gate and pass gateId. Visibility overrides belong in advanced client.verify() flows.

​

Never ship publicly

Raw signatures, owner-only payloads, private proof enumeration.

​

Smart contract audit

NEUS smart contracts were audited by SafeStack AI in March 2026. The audit completed with no critical, high, medium, or low severity findings. Scope: NEUSVerifierRegistry.sol, NEUSVoucherHub.sol, NEUSVoucherSpoke.sol, NEUSToken.sol. Public status is listed on the Trust Center. Contact dev@neus.network for the full report.

​

Diligence material

​

Checklist

​

Next