Policy & Access
Zero-credential execution for AI agents
To perform complex tasks, AI agents need access to databases, APIs, and administrative tools. Handing an autonomous AI standing credentials to all of them is exactly what security review will not allow. NuDay keeps the agent separate from its identity, so agents use access without ever holding it.
NuDay separates the agent from its identity, and from the user's identity. When work is delegated on-behalf-of (OBO) a user or another agent:
Only one credential travels, the agent's or the user's, never both.
The risk
The danger of credentialed agents
Most orchestration platforms and security vendors inject OAuth tokens, API keys, or session credentials directly into the agent's runtime or context window. This creates an unpatchable vulnerability:
Memory leaks and storage
LLMs are designed to remember context. Hand an agent a token and it can memorize it, write it to plain-text conversation logs, or store it in its RAG history.
Replay attacks
If an attacker compromises the agent via prompt injection, they can extract those stored credentials and execute devastating replay attacks against your data and infrastructure.
The approach
On-behalf-of execution
The agent is completely separated from your secrets. A dedicated NuDay Policy Server brokers all access.
Zero-knowledge architecture
When an agent needs a customer record from Salesforce or a query in Postgres, it does not use a credential. It sends a cryptographic request to the Policy Server.
Secure brokering
The Policy Server verifies the agent's identity, checks enterprise policy, mints a short-lived ephemeral credential scoped to that single action, and executes the call. Nothing standing is ever handed out.
Data, not keys
The Policy Server returns only the requested data back to the agent. The agent never sees the API key or token, so a hijacked agent has no credential to steal or replay.
NuDay ensures the AI agent never handles authorization credentials, preventing replay attacks and data exfiltration.
Human oversight
Real-time human-in-the-loop approvals
Autonomy must be governed by human oversight. NuDay enforces strict boundary conditions for high-risk actions, such as initiating a wire transfer, modifying production code, or accessing protected health information.
Dynamic access requests
If an agent attempts an action beyond its baseline permissions, the Policy Server pauses execution and triggers a real-time access request.
Third-party approvers
The request is instantly routed to a designated human approver through your existing workflows: Slack, Microsoft Teams, or email.
Contextual escalation
The approver sees exactly what the agent wants to do, why, and on whose behalf. Once approved, the Policy Server completes the OBO execution.
Integration
Seamless enterprise IAM and PAM integration
NuDay does not replace your enterprise identity stack; it extends your existing Zero-Trust architecture to your AI fleet.
Directory sync
Native integration with LDAP, Active Directory, and OIDC, so an agent's permissions are dynamically tied to the human user's organizational role.
PAM compatibility
Integrates with existing Privileged Access Management solutions. For a privileged task, the Policy Server checks out a short-lived, ephemeral credential from your PAM vault.
Unified policy enforcement
Manage agent access policies from the same centralized console you use for human access, for consistent governance across the enterprise.
Comparison
Agent access models, side by side
| Feature / risk profile | NuDay (Policy Server) | Legacy orchestrators | AI "firewalls" |
|---|---|---|---|
| Where credentials live | Isolated in the Policy Server | Injected into agent context | Injected into agent context |
| Risk of agent memorizing keys | Zero (agent never sees the key) | High (often logged in plaintext) | High (firewalls only filter outputs) |
| Vulnerability to replay attacks | Zero (tokens never exposed) | High | High |
| Execution model | On-behalf-of (OBO) execution | Direct agent execution | Direct agent execution |
| Native human-in-the-loop | Real-time approval routing | Requires custom coding | Output blocking only |
Eliminate AI credential risk
Zero-credential execution, with real-time human oversight.
See how NuDay's Policy Server brokers every agent action without ever handing out a standing credential.