getai.id vocabulary,
defined.

The identity layer for AI agents. Built and operated by Deep Identity Inc., powered by the open UAIIP protocol.

getai.id is the product. It binds every AI agent to a biometrically-verified human operator via a soulbound on-chain credential, and lets relying parties verify identity through zero-knowledge attestations — confirming verification status and regulatory attribution without exposing any personal data. The cryptographic protocol underneath is UAIIP.

Universal AI Identity Protocol — the open cryptographic protocol that powers getai.id.

UAIIP defines the on-chain registration model, the soulbound credential format, the SHA-3-512 model fingerprinting algorithm, and the UAIIP-ATTEST-v1 attestation envelope. The specification, SDKs, and smart contracts are open source. getai.id is the hosted, commercial implementation built and operated by Deep Identity Inc.

Agent Soulbound Identifier — a 256-bit unique identifier minted as a non-transferable ERC-721 token on an EVM L2.

An ASID is the canonical, globally-unique identity for a registered AI agent. It contains references to the operator's VOC, the agent's model fingerprint, a packed capability vector, and jurisdictional declarations. ASIDs cannot be transferred — the smart contract reverts on any transfer attempt, preventing identity laundering.

The biometrically-anchored credential issued to a verified human operator, signed by deepidv's certificate authority.

A VOC attests that a specific natural person or legal entity has been verified to a defined assurance level via document validation, biometric match, liveness detection, and sanctions screening. The VOC hash is recorded on-chain; the underlying biometric data never leaves deepidv's secure infrastructure.

A non-transferable ERC-721 variant — transfers revert; the token cannot be sold, gifted, or reassigned.

Soulbound tokens enforce non-transferability at the smart contract level. UAIIP uses a modified ERC-721 that overrides the transfer function to revert. Acquisition of an AI system mints a fresh token through a new VOC verification flow, preserving an auditable chain of custody.

A cryptographic proof that an operator is verified without exposing any identifying data.

Zero-knowledge attestations let an AI agent prove to a relying party that its operator has been verified, that specific regulatory frameworks are satisfied, and that the agent is authorized for a given capability — all without transmitting the operator's name, document, biometric, or any other PII. The verifier receives a yes/no plus regulatory attribution.

A statistical signature of an AI model's outputs used to attribute unattested content to a registered system.

Generated from a standardized prompt battery and the model's response distribution. Behavioral fingerprints are tolerant of inference-time variation (temperature, top-k) and let UAIIP perform passive detection of unregistered or modified agents in the wild via deepcam-derived analysis.

A SHA-3-512 cryptographic hash that uniquely identifies a specific AI model based on its weights and configuration.

Computed deterministically over the canonicalized weight tensors, normalized architecture graph, and inference configuration metadata. Fine-tuned models reference their base model's fingerprint plus the delta, building an unbroken provenance chain for every registered system.

A packed uint256 with bitwise capability flags declaring an agent's authorized functional scope.

Each bit position corresponds to a defined capability class — text generation, financial transactions, autonomous decision-making, medical diagnosis, etc. Compliance contracts can check whether a registered system has declared a specific capability with a single bitwise operation.

The unbroken provenance graph linking a fine-tuned or merged model to all of its parent fingerprints.

Every fine-tune (DPO, RLHF, SFT, continued pre-training) and every merge (SLERP, TIES, DARE) records its method and parameters alongside the prior fingerprint, so any registered model can be traced back to its open-weight ancestor or proprietary origin.

The active attestation protocol — HTTP header for APIs, token biasing for text generation, steganography for media.

UAIIP-ATTEST-v1 specifies how a registered agent embeds a cryptographic proof of identity into its outputs. APIs include the X-UAIIP-Attestation header. Text outputs encode the proof through statistically imperceptible token biasing. Images, audio, and video use media-type-specific steganography.

deepidv's deepfake detection product — the multi-modal analysis engine UAIIP uses for passive agent attribution.

deepcam analyzes images, video, and audio for AI-generation artifacts including GAN frequency-domain anomalies, diffusion residuals, and synthetic-speech markers. UAIIP uses deepcam to score outputs against the behavioral fingerprints of registered models, attributing unattested content with calibrated confidence.

A real-time 0–100 reputation score derived from behavior, age, denial frequency, and scope breadth.

Trust Scores are publicly verifiable but reveal nothing personal. They compress an agent's full registry history (anomalies, revocations, attestation health) into a single metric a relying party can use as a precondition for sensitive actions.

One of seven hierarchical authorities an agent can be granted: read, write, execute, transact, communicate, delegate, elevate.

Scopes can carry time windows, rate limits, and conditional constraints. Compliance contracts and Arbiter (deepidv's risk engine) check scope membership at sub-5ms latency through Redis-cached permission verification.

The Ethereum-compatible Layer 2 chain on which UAIIP soulbound tokens are minted.

Selected for low transaction costs, high throughput, and institutional credibility. Sensitive metadata too large for on-chain publication is stored on IPFS or Arweave with the content hash recorded in the token, balancing transparency with confidentiality.

A modified ERC-721 token standard that reverts on every transfer operation, enforcing non-transferability.

UAIIP's AIIdentityRegistry contract overrides every ERC-721 transfer entry point to revert the transaction. Indexers and wallets still recognize the asset because the rest of the standard is untouched; only the transfer pathway is closed off.

deepidv's AI risk-management module — sanctions, PEP, adverse-media, and behavioral anomaly screening for registered operators.

Arbiter is the compliance engine plugged into the UAIIP registry. Every registered operator is screened in real time against authoritative sanctions lists, PEP databases, and adverse-media feeds; behavioral anomaly detection flags drift from declared capabilities.