One permission model for every request
Ownership, roles, teams, and direct grants combine into one effective permission, inherited down your organization tree. Conditions on time, network, location, and device layer on top, enforcing or report-only.
Security & compliance
Security and compliance, built into the platform
Neuraforge runs every person, AI agent, and automation through one identity model, one policy layer, and one audit trail. Here is how the platform protects your data and helps you evidence the frameworks you answer to.
Identity and access
How people, agents, and machines prove who they are and what they may do.
Single sign-on and provisioning
Per-organization OIDC and SAML 2.0 single sign-on, SCIM 2.0 provisioning, and directory sync. A removal safety brake holds large changes for approval, and a disabled account loses access automatically.
Strong sign-in, live sessions
Authenticator apps, one-time codes, backup codes, and passkeys. Sign-in policies restrict by network, time window, device, browser, and location, and live sessions are re-checked when conditions change.
Reviews, requests, and just-in-time access
Time-scoped access reviews revoke on lapse, access requests go to approvers, and an effective-access report answers who holds what, by which path. Machine identities get their own credentials and audit trail.
Data protection and encryption
How customer data is separated, sealed, and erased.
Per-organization encryption keys
Secrets and credentials are encrypted with keys unique to each organization and wrapped by your choice of key service, including your own. Rotating the key re-encrypts what it protects.
Isolation between organizations
Every root organization is a hard data boundary. A child organization can exclude its parents from decrypting its data, so a parent-side leak cannot expose it.
Secrets managed once, never copied
Ledger stores secrets in one governed vault. Products reference a secret by name, every read is logged, version history keeps no readable value, and rotation updates every consumer at once.
Erasure by key
When an organization is terminated, its encryption keys are destroyed after a grace window, leaving its sealed data permanently unreadable. Retention and data-residency settings bind every AI call.
AI-specific safeguards
Controls built for models, agents, and the tools they call.
One governed path for every model call
The gateway applies authentication, guards, routing, budgets, and limits in one pipeline, and writes an audit row for every call, including the ones it refuses.
Personal data redacted in flight
Shroud detects personal data in text, documents, and images and redacts it before a request reaches a provider. Recovering a protected value takes permission and leaves an audit record.
Agents as accountable members
Each AI agent holds roles in the same access engine as your people, with its own revocable credential and cost caps. One switch stops the fleet, and risky acts need a person's approval.
Defense against hostile prompts and tools
Vigil screens prompts and completions for injection and jailbreak attempts, scans tool servers for hidden instructions, quarantines what fails, and reconciles declared AI against what is actually in use.
Audit and evidence
Records you can verify and evidence you can hand to an assessor.
A tamper-evident audit trail
Platform actions are chained per organization, so any change to, removal of, or reordering of an audit row is detectable by recomputing the chain. A daily checkpoint can be sent to your own systems.
Decisions that explain themselves
Apex keeps an append-only decision ledger, so you can ask why an action was blocked and get the policy and version behind it. Refused calls and actions carry a typed denial and a decision reference.
Continuous compliance evidence
Oath collects evidence from the platform's own records, scores readiness, and explains each gap. Signed audit packages verify offline, and auditors get time-boxed, read-only access.
A trust page you control
Publish a validated snapshot of your posture on a public trust page. Gated documents need NDA acknowledgement and an approved, time-limited grant, and every download is audited.
Change control and operations
How change, devices, and remote access stay governed.
Approved, versioned change
Saga records who changed what, with side-by-side diffs and rollback that creates a new version. Approval workflows sit in front of production, and running systems always use the approved version.
Decide before it happens
A single pre-effect gate scores sensitive acts such as secret reads, access changes, and device commands. Decision policies can notice, warn, or block, and you can replay a draft policy on past activity first.
Drift detection and guarded rollout
Saga re-verifies live configuration against its recorded history and captures out-of-band changes as attributed versions. After a change applies, a spike in errors offers a rollback a person confirms.
Governed endpoints and remote access
Ember connects outbound only, with no inbound ports, and measures disk encryption, firewall, and update health. Wipes, locks, and commands can require a second person's approval.
Evidence for the frameworks you answer to
- NIST 800-53
- SOC 2
- ISO 27001
- HIPAA Security Rule
- CSA CAIQ
- NIST CSF 2.0
- FERPA
- GLBA
Oath's built-in control packs help you collect evidence and measure readiness against these frameworks. They support your own audits and assessments; they are not a certification or attestation.
Have a security review coming up?
Tell us what your reviewers need to see and we will walk you through the detail behind any section on this page.
Talk to us