Data
Ohalo
Data X-Ray finds sensitive files across your systems and labels them. Virtru reads those labels and encrypts each file automatically, before it moves.
Cryptographic proof that a human approved what the agent is about to do.
Dapple captures a signed human approval for each AI agent request. Data Harbor Platform enforces exactly that approval on every data access.
How Dapple Security and Data Harbor Platform work together, and what each one does.
Dapple Security is an identity security platform built around a simple belief: digital access, identity, and security belong in the hands of the human who owns them. Dapple replaces passwords with biometric, phishing-resistant authentication, issuing channel-first software passkeys that work across the devices and identity systems an organization already runs. As a FIDO Alliance member, Dapple is built to the same open standards driving the industry’s move away from passwords and toward verifiable, hardware-backed human identity.
That same discipline, a real person, cryptographically verified, making a real decision, is what helps Dapple solve one of the hardest problems in enterprise AI: governing what autonomous agents are allowed to do on a person’s behalf. Dapple Identity authenticates the analyst through the organization’s own identity services and human public key infrastructure, including hardware-backed smart-card certificates, then puts each AI agent’s specific request in front of that person in plain language. Approval isn’t a checkbox; it’s a signed, time-bound grant that binds together the identity of the person, the agent acting for them, the purpose of the request, the exact permissions involved, and how long that authority lasts. Any attempt to alter an approval after the fact invalidates it outright.
How it works together
Dapple’s signed approval is the first part of agentic AI governance: a real human, looking at a real request, saying yes with their own cryptographically verifiable identity. The Virtru Data Harbor Platform extends that governance by treating Dapple’s signed grant as the first of two credentials required on every protected request. The second is an independently verified identity for the specific AI agent workload attempting the access. When an analyst approves a request through Dapple, the platform validates that signed grant and derives a short-lived access token scoped to exactly one verified agent. The token inherits the grant’s limits exactly, it cannot outlive the grant’s expiration, and it cannot carry more permission than the analyst actually approved. Virtru then enforces that delegated scope on every model call, tool call, and touch of protected data for as long as the grant remains valid. The analyst’s decision and its enforcement come from one integrated design, not two systems hoping to agree.
Where they connect
Dapple Identity issues a signed, time-bound grant through human PKI that links a specific analyst, agent, purpose, and permission set into one tamper-evident record. The Virtru Data Harbor Platform validates that grant and derives a short-lived access token scoped to one attested agent workload, a token that can never outlive or widen what was actually approved. Every AI agent action traces back to one human’s explicit, revocable decision, not a standing credential an agent could carry indefinitely.
Dapple shows the analyst exactly what an agent is requesting and captures explicit approval before any data access occurs. That approval becomes one of two credentials the Data Harbor Platform requires on every protected request, the other is the agent’s own attested workload identity. Agents start with zero access to protected content and can only reach the data a specific human chose, for the task that human chose, for only as long as that human chose.
Dapple’s signed grant defines the full scope of what an analyst is willing to delegate. Virtru’s Data Cleanroom enforces that a main agent can only hand sub-agents equal or lesser access, never broader or longer, and revoking the parent grant cuts off every descendant agent instantly. Delegation sprawl stays bounded no matter how many agents a single approval sets in motion.
When an analyst revokes a grant, access stops going forward immediately, without touching the analyst’s own underlying entitlements or undoing work the agent already completed. That combination, a real human decision, cryptographically signed, enforced down to a single derived token with a hard ceiling, is why this architecture is drawing early interest from federal defense and intelligence programs under executive mandate to deploy AI agents without losing Zero Trust guarantees in the process.
What you get when Dapple and Data Harbor Platform run side by side.
Confirms the analyst’s identity with biometric, phishing-resistant authentication before any delegation begins, then shows that analyst each AI agent request in plain language for explicit approval.
Issues approvals as signed, time-bound grants through human PKI, including hardware-backed smart-card certificates, that bind the user, agent, purpose, permissions, and duration into one tamper-evident record; any altered approval is rejected outright.
Never acts on a signed grant alone, the Virtru Data Harbor Platform also verifies the requesting agent’s attested workload identity, so identity and permission are evaluated independently before any data moves.
Derives a short-lived access token from each signed grant, scoped to one verified agent workload; the token can never outlive the grant’s expiration or exceed the permissions the analyst actually approved.
Lets an agent learn whether relevant data exists, and roughly how much, before ever requesting access, so an analyst approves a narrow, specific scope instead of a broad one.
Allows a main agent to hand sub-agents equal or lesser access only, never broader or longer, with revocation of the parent grant instantly cutting off every sub-agent working under it.
Ties every request, approval, and access event to the specific analyst, agent, and approved permission in a single trail, supporting the oversight defense, intelligence, and regulated enterprise programs require.
Deploying Dapple Identity behind the Virtru Data Harbor Platform to govern signed, human-approved delegation for AI agents requires active deployments of both platforms, a Non-Person Certificate Authority for agent workload identity, and the identity and entitlement services both platforms integrate against.
Integrated with the customer’s identity services and human PKI (e.g., smart-card or other hardware-backed certificates) to authenticate analysts and capture explicit, signed approval.
Customer VPC, on-premises, or air-gapped; validates signed grants and issues scoped, short-lived access tokens.
Issues and validates workload identity certificates for AI agents (a SPIFFE/SPIRE-based workload identity architecture is supported).
The customer’s existing entitlement system(s) supplying the attributes that define each analyst’s delegable authority.
Databases, vector stores, object storage, or code repositories, each fronted by a Virtru resource connector or policy enforcement point.
Dapple and Virtru extension points for approval prompts, scope display, and audit navigation, integrated into the customer’s existing agent or analyst interface.
How the two platforms split the work across security, compliance and privacy.
Dapple and the Virtru Data Harbor Platform operate at complementary layers of one architecture, not two bolted-together products. Dapple Identity governs the human decision: it authenticates the analyst, presents the specific agent request, and captures a cryptographically signed, time-bound approval. The Virtru Data Harbor Platform governs everything that happens after that decision, validating the signed grant, deriving a workload-scoped token that can never outlive or widen it, and enforcing that boundary on every model call, tool call, and data access an agent attempts. Two credentials travel with every protected request: the analyst’s signed grant and the agent’s attested workload identity. Both must be independently satisfied before any data moves, and agents start with no access to protected content by default.
Compliance
The Virtru Data Harbor Platform is built on the OpenTDF open standard, is FedRAMP Moderate authorized, and holds SOC 2 Type II. The joint architecture is designed around the dual-credential model Zero Trust frameworks call for: a signed human approval and an independently verified agent workload identity, evaluated together before any data access, with workload identity anchored to a customer-controlled Non-Person Certificate Authority.
Privacy
Decryption authority and entitlement data stay where they already live. Dapple authenticates the analyst and issues a signed approval tied to a specific, narrow scope, it does not hold the customer’s protected data. The Virtru Data Harbor Platform evaluates that approval against the customer’s own authoritative entitlements and policy, and releases access only when both credentials are satisfied, so neither party needs to hold the underlying data or keys outside the customer’s control. Governed metadata discovery lets an agent learn whether relevant data exists, and roughly how much, without exposing content, titles, or record identifiers before approval is granted.
Data
Data X-Ray finds sensitive files across your systems and labels them. Virtru reads those labels and encrypts each file automatically, before it moves.
Infrastructure
Encryption keys stay in Thales hardware-backed storage, under your control. Virtru protects the data itself, so it stays secure wherever it goes.
Infrastructure
Access control for every object, not every bucket. Virtru applies policy to each file stored in Amazon S3, so teams stop creating buckets to share safely.