Security · governed AI infrastructure

Every AI request is authorised before it runs

Identity, policy, and scope verified at the gateway, before any token reaches a model. Nothing granted by default, everything expires, all decisions logged.

No standing credentials
Every request attributable and revocable
01 · Security foundation

One request.
Protected at every step.

Every AI request receives scoped access, passes policy, and leaves an auditable trail.

Developer

developer@company.com 
await client.messages.create({
  model: "claude-sonnet-4-5"
})
01

Identity

Verify who is making the request.

02

Scope

Mint a request-scoped credential.

TTL 04:00
AIMS Gateway Request verified
  • developer@company.com
  • project-x / cart-flow
  • Claude Sonnet 4.5
TTL 03:42
Policy Passed
Entitlement Allowed
Credential
  1. Active
  2. Expiring
  3. Revoked
03

Policy

Check entitlement and risk in real time.

Policy passed
04

Access

Allow the request with short-lived credentials.

AI Provider

Request executed with scoped access.
  • Claude Sonnet 4.5
  • Response generated
  • No provider key exposed
Provider key never exposed

logged · req_7f3a91

  • Short-lived credentialsAuto-expire after a limited time.
  • Central revocationInstantly kill from the control plane.
  • Full audit trailEvery request is logged.
  • Provider keys stay safeNo long-lived keys exposed.
02 · Security model

Every request passes through a security boundary

Nothing reaches a model without first clearing identity, authentication, and policy. The gateway is the only path in. Here's what happens to a single request, in order.

Live request
watching the pipeline

Event stream
03 · Identity

Know who is making every request.

Every request arrives with a resolved identity and project context. AIMS validates the session before policy evaluates what that identity is allowed to do.

Developer

Makes an AI request

Identity
developer@company.com
Project
checkout-redesign
Client
VS Code
Session
8F2A-91C7
Request
Claude Sonnet 4.5
AIMS Identity Identity verified
  1. 01

    Identity lookup

    Resolve user and team

  2. 02

    Session validated

    Verify session token

  3. 03

    Context resolved

    Load project and entitlements

  4. 04

    Identity verified

    Ready for policy evaluation

Resolved by AIMS

Verified UTC

  • Payments squadTeam
  • 3 modelsEntitlements
  • Within limitProject budget
  • DevelopmentEnvironment
Policy evaluation Ready . Go to 04, policy enforcement. Identity verified before policy enforcement.
04 · Policy enforcement

The model only sees what policy allows

Every authenticated request is evaluated at the gateway. AIMS checks identity, project, entitlement, model, feature, budget, and environment before a provider ever receives the request.

Request

From developer · via VS Code

Identity verified in 03
Actor
developer@company.com
Project
checkout-redesign
Feature
code-completion
Client
VS Code
Model
Claude Sonnet 4.5
Payload
inline · 3.1k tok

AIMS Policy Gate

Evaluating against your organisation’s policies

Preview outcome
  1. 01 IdentityVerified user and session token Allow
  2. 02 ProjectMember of checkout-redesign Allow
  3. 03 EntitlementRole grants governed model access Allow
  4. 04 ModelApproved for this project Allow
  5. 05 Featurecode-completion is enabled Allow
  6. 06 Budget82% of budget used Allow
  7. 07 EnvironmentProduction gateway, approved region Allow
Access granted Request forwarded to the provider 7 / 7 passed
Model

Claude Sonnet 4.5

Executed
  • Request received
  • Inference executed
  • Response generated
  • No provider key exposed

Request never reached provider.

Latency
1.2 ms
Ruleset
v3.14
Matched rule
default-allow
Scope
feature-x
05 · Credential security

No long-lived keys.
Access exists only when it is needed.

Every call receives scoped access for a limited window, then the credential expires automatically.

Credential lifecycle: each request receives a credential scoped to project checkout-redesign, feature-x, and model GPT-5. It is valid for 5 minutes 22 seconds, is never written to disk, and is revoked automatically when the window closes, or earlier from the central console.

  1. 01

    Requested

    Developer request received

  2. 02

    Issued

    Short-lived credential created

AIMS access pass
Project
checkout-redesign
Scope
feature-x
Model
GPT-5
Credential automatically revoked.
Active
03 Active Credential valid for the request window

  1. 03
    Active

    Credential valid for the request window

  2. 04

    Expiring

    Approaching TTL limit

  3. 05

    Expired

    Credential automatically revoked

  • Never storedNever written to disk.
  • Request scopedLimited to project, feature, and model.
  • Automatically expiresValid only for the required time window.
  • Centrally revocableCan be revoked at any time.

Access exists only for the work that needs it.

06 · Data protection

Your data stays inside a
governed boundary.

Every request is encrypted in transit and passes through AIMS before reaching an AI provider, with no direct, unmanaged connections.

Developer

Makes a request
User
developer@company.com
Project
checkout-redesign
Model
GPT-5
Payload
3.1k tokens

Response received · via AIMS

AIMS Gateway

Secures · Governs · Routes
  • TLS 1.3 encryption
  • Policy enforcement
  • Provider routing
Governed boundary No direct provider connection

AI Provider

Processes the request
GPT-5 · Azure OpenAIComplete
  • Receives encrypted request
  • Processes within governed path
  • Returns encrypted response
  • No direct developer connection

  • Encrypted in transit

    All data is encrypted using TLS 1.3.

  • Governed by AIMS

    Project, model, feature, and permissions are enforced.

  • No direct provider access

    Traffic always routes through AIMS.

  • Protected response

    The response returns through the same governed path.

07 · Auditability

Every decision leaves evidence

Every request, policy decision, and credential event is written to the audit trail, allowed and denied alike, and linked by a shared request ID.

Audit event stream Live record of all requests, policy decisions, and credential events.
Live streaming
View:
Tracing

No matching requests in the current window.

Select an event to see its record · click a request ID to trace it.Tap an event to see its record · tap a request ID to trace it. Illustrative events
08 · Security control plane

A live view of the system that enforces everything above

Identity, policy, credentials, the provider gateway, and the audit pipeline are independent services. The console below reflects their status as they run.

Security control plane AIMS Console
Operational
  1. Identity Verifies access Session and project context resolved firstSection 03
  2. Policy engine Enforces rules Seven checks on every requestSection 04
  3. Credential service Issues short-lived access Scoped per request, revoked on expirySection 05
  4. Provider gateway Routes governed requests Provider keys never leave the gatewaySection 06
  5. Audit pipeline Records every decision Allowed and denied, linked by request IDSection 07
Requests (today)
128,423Total requests
126,930Authorised
1,493Blocked
38Active credentials
Request volume Live
Recent events View all
  • 09:42:12 Credential TTL 14m · feature-x Issued
  • 09:41:58 Policy All checks passed (7/7) Allow
  • 09:41:33 Policy Budget limit exceeded Blocked
09 · Security principles

Every request leaves a trail you can defend

AIMS turns AI access into an explicit decision: who requested it, what they were allowed to use, how long access lasted, and what happened next.

  1. Default deny

    No entitlement, no request. A developer cannot reach a model or provider until the project explicitly allows it.

  2. Short-lived access

    Every grant has an ending. Credentials are scoped to the task and expire when the work no longer needs them.

  3. Attributable usage

    Identity stays attached. Requests retain their developer, project, story, provider, and authorisation context.

  4. Evidence by default

    Allowed and denied are both recorded. The audit trail captures the decision, not only the successful request.

REQUEST LEDGER req_7F2A

AUTHORISED Scoped access issued after policy approval.
Actor
Maya Chen / platform team
Context
checkout-api / payment retry story
Provider
Anthropic / Claude Sonnet
Decision
Policy matched / entitlement confirmed
Credential
Scoped to checkout-api / expires in 03:58:21
Audit
Decision recorded / 09:42:18 UTC

AIMS · Security

Every request accounted for. Every decision defensible.

Request early access Explore AIMS