Microsoft Entra
Identity & directory
Operator sign-in via OIDC. Not an agent action path; without an IdP configured, operators use a client-set bearer token.
The governance layer for enterprise AI agents.
Export customer records
Read and export 2,400 customer records from a CRM through crm.contacts.export.
Approval applies to this request only, with approvers chosen by your policy.
| Agent | Committed | Ceiling | Utilization |
|---|---|---|---|
| analytics_01 | $10.00 | $50.00 | |
| support_02 | $3.20 | $20.00 | |
| data_04 | $44.50 | $45.00 | |
| research_07 | $12.00 | $30.00 |
Illustrative aggregation of work-budget commitments by agent. This display is not a vendor billing meter; committed amounts are authorization charged against configured decision-time ceilings, not a measurement or guarantee of final vendor charges.
analytics_01crm.contacts.export · 2,400 recordsPriya RamanAuthorized#4181 · 77b1c0e2…03aa9ab42a…0c91Designed for the stack you already use
of agentic AI projects will be canceled by the end of 2027, a Gartner forecast rather than an observed outcome, citing escalating costs, unclear business value, inadequate risk controls.
Gartner, press release forecast, 2025higher average breach cost for organisations with high levels of shadow AI, in IBM's 2025 study of 600+ breached organisations. 97% of those that suffered an AI-related breach lacked proper AI access controls (an association in that study population).
IBM, Cost of a Data Breach, 2025machine and AI identities per human identity, and 61% of privileged access requests still fulfilled with standing privilege. The figures come from Palo Alto Networks' 2026 Identity Security Landscape Report, a survey of 2,930 cybersecurity decision-makers.
Palo Alto Networks, Identity Security Landscape Report, 20261 IBM, Cost of a Data Breach 2025: 87% of the studied organisations reported no governance policies or processes to mitigate AI risk (an association in that study population, not a causal measure).
AI agents need access to your applications, data, and systems to be useful. Ungoverned, that access can read, change, spend, or delete a whole database, without anyone approving the action. AAES is pre-launch; the controls below describe intended behavior, not generally available functionality. For supported enforced paths, AAES is being designed to deny dispatch or withhold the authority the action needs. It would not control credentials or execution routes outside that boundary. On supported observed paths, AAES would record the telemetry it receives but could not block the action.
Each request checked
through AAES.
Accountable employees approve
when policy requires it.
Planned accounting rule: each approved request would reserve its amount against the configured budget, and reservations are not released until the work item closes. This caps commitments recorded through AAES, not final vendor charges or spending outside AAES.
Planned for enforced paths: one signed record per action, and AAES would refuse dispatch if it cannot durably record the decision. Observed paths cannot be blocked; their records may be incomplete.
Check agent requests routed through AAES against permissions, approval requirements, and spending limits before execution.
Keep employee access in your existing identity provider.
How will you set up approval routing?
Agent useChecks before action
Named approvers
Named in AAES.
No directory import.
Check each request
Keys in your vault.
Requests through AAES.
Employee accessOutside AAES
AAES can block an action when it controls the required credentials and the agent cannot bypass that path.Observation-only is not enforcement.
The agent names the capability it wants to use. On the enforced path, the destination and credential come from the registration, never from the agent.
analytics_01crm.contacts.export2,400 recordsEvery registered agent has at least one accountable human manager. Name the approvers in AAES.
The approval policy determines who must decide each request.
Determines who must decide.
Know which agent is acting and who is accountable.
Through AAES, agents only see the capabilities they are permitted to use. Anything else is a request for approval, not a grant of employee access.
On an enforced path, a capability the operator registers as irreversible always requires approval by an authorized person. No policy can switch that off, and an agent can never approve itself. Each action approval covers exactly one request, exactly as reviewed, and it expires. AAES does not automatically recognize irreversibility; the operator’s registration and classification are declarations.
On requests through AAES, the amount is committed when the decision is made.
Over the ceiling, the request is blocked. Other vendor bills such as LLM tokens and cloud are not capped here.
| Agent | Committed | Ceiling | Utilization | Decision |
|---|---|---|---|---|
| analytics_01 | $10.00 | $50.00 | Within budget | |
| support_02 | $3.20 | $20.00 | Within budget | |
| data_04 | $44.50 | $45.00 | Near ceiling | |
| research_07 | $12.00 | $30.00 | Within budget |
Requirements for AI agents differ across these nine jurisdictions. AAES produces records to support compliance reviews, showing how it brokers credentials on enforced paths, requires human approval for irreversible actions, and applies spending limits.
Start with your jurisdiction.
Evidence supports compliance review, not a compliance determination.
Keep your identity provider, collaboration tools, vault, and monitoring. AAES connects them to agent permissions, approvals, and action records.
Identity & directory
Operator sign-in via OIDC. Not an agent action path; without an IdP configured, operators use a client-set bearer token.
Identity & directory
Operator sign-in via OIDC. Not an agent action path; without an IdP configured, operators use a client-set bearer token.
Directory & collaboration
Directory and surface registration. An agent holding its own workspace credential bypasses AAES; that path is observational at best.
Enterprise communication
Capability pack, enforced: the bot token stays in your vault and AAES injects it per call. Bypass: a token issued directly to the agent.
Secrets & key custody
The enforced path's credential source: AAES holds a reference and fetches at use. Bypass: material distributed outside the referenced path.
Evidence & monitoring
Record export for your monitoring. Observational: it receives evidence and prevents nothing.
Tested against simulated vendor APIs. Compatibility with your live environment is validated during the design partnership.
Tell us which tools your agents need to use. We’ll confirm what AAES supports, explain any limitations, and define what needs validation in your environment.
Deploy on your infrastructure and control your keys, storage,
and availability. Start on a single VM or deploy in your enterprise environment.
Run AAES with Docker Compose. Records live on your host. You create and control the keys.
You operate keys, storage, and availability; no availability SLA is offered.
Check exported records for integrity without connecting to an AAES service.
Read the Evidence Libraryanalytics_01crm.contacts.export · 2,400 recordsPriya Ramanallow_export · R3$10.00 / $50.00Enforcede94d…f7c077b1…03aa3f9c…8e21Check offline that the sealed record, including its authorization, scope, and named approver, has not changed. Use a public key you already trust or verify separately. A key included with the export does not, by itself, prove where the record came from.
A configured independent witness or timestamp authority can provide external evidence about a signed head, within that service’s scope. Neither is connected by default. That evidence does not attest to capture completeness, policy correctness, enforcement effectiveness, or regulatory compliance. Work outside mediated paths is not captured; caller-submitted reports are recorded as claims, not verified execution. A recorded dispatch does not prove the outcome, and observation-only records do not establish enforcement.
AAES (Autonomous Agentic Enterprise Systems) is the governance layer for enterprise AI agents. It applies permissions, approvals by authorized persons, and spending limits to actions routed through AAES, and records each decision. An action whose decision cannot be recorded does not proceed. Each registered agent has at least one accountable human manager. Employee access stays in your existing identity provider.
No. Keep employee access in Microsoft Entra, Okta, or your existing identity provider. Name the agent’s first approver in AAES. A directory integration can supply additional approvers where required; live tenant compatibility is validated during the design partnership.
On an enforced path, a capability the operator registers as irreversible always requires approval by an authorized person; no policy can switch that off. Policies can require approval for other requests too. Each approval covers exactly one request: the intent and the exact request content the approver reviewed. If anything in the request changes, the approval no longer applies, and every approval expires. Actions through alternate paths are not prevented by this. An action approval applies only to the bound request; an access approval instead authorizes the specific entitlement requested, which may permit multiple requests within its scope and lifetime, each still subject to its action-level controls. The API also records agent-manager decisions and automated approval paths; these are distinct from approvals by authorized persons. They can never satisfy the mandatory authorized-person approval for a capability registered as irreversible, and an agent cannot approve its own request. A named approver recorded on a record is an attribution field; it does not by itself establish that the person was authenticated and authorized. The approver is the person authorized to decide that request; an agent cannot approve itself.
Yes. A capability can be registered as observed, with every record labeled accordingly. AAES can block an action when it controls the required credentials and the agent cannot bypass that path. While the agent retains its own direct credentials, AAES can record reported activity but cannot stop the call.
The agent authenticates to AAES and requests a registered capability. AAES evaluates policy, obtains any required approval, journals the decision, and executes the permitted call using the configured customer-controlled credential path. The broker returns the permitted result rather than handing the downstream credential to the agent. This describes brokered execution, not observation-only integrations. Credential exposure through downstream responses, errors, logs and records remains a deployment test requirement. Credential isolation is not data isolation: AAES does not control arbitrary subsequent use of data already delivered to an agent.
Requests requiring a new AAES decision fail closed if AAES is unavailable or the required decision journal cannot be written: no new grant or brokered execution is authorized through that path. Previously issued grants can remain usable until expiry, for up to 15 minutes. This does not cancel work already authorized or undo downstream effects. Observation-only and bypass paths are not stopped by AAES’s unavailability. Employee access remains governed by your identity provider. You operate availability on your infrastructure; no availability SLA is offered today.
You retain your exported records and the offline verifier supplied with your deployment. Check exported records for integrity without connecting to an AAES service. A configured independent witness or timestamp authority can provide external evidence about a signed head, within that service’s scope. Neither is connected by default. That evidence does not attest to capture completeness, policy correctness, enforcement effectiveness, or regulatory compliance.
If your agents live on one vendor’s cloud and that vendor’s tooling satisfies your reviewers, those tools are a reasonable answer. AAES is designed for estates that need customer-controlled custody of third-party credentials, approval as a precondition of dispatch, per-action ceilings on business spend through your own payment credentials, and exported records designed for offline integrity verification, across agents on any platform. AAES is pre-launch; the differences are stated as claims with dated sources on the comparison page, which also credits what the vendors genuinely offer.
Bring one real agent workflow, a named operator, and weekly feedback.
Here is what we bring to the design partnership.
We test native adapters against your live tenants during the design partnership. You get a coverage matrix showing what works and where gaps remain.
Help set what gets built next. Your agent workflow becomes a reference case we test the product against.
The design-partner license is $40,000 for 6 months. It includes a named engineer and a direct channel to the AAES team.
Start with a short introduction about your workflow and stack.

Agree on the approval, evidence, and acceptance criteria before a design partnership.
Now inviting design partners.