1. Posture at a glance
AAES is the governance layer for enterprise AI agents: it decides whether a proposed agent action is authorized, brokers the credential at the moment of use where configured, and seals a record the client can verify offline. This page states what is verified, what is planned, and what does not exist. Where a page and the code disagree, the code is authoritative and we correct the page.
Customer-run evaluations by request. No production adoption claimed.
Runs in your environment. No AAES hosted service today.
No SOC 2 report exists today. Hosted offering gated on Type I.
Scope and plan exist; engagement not yet executed.
2. Compliance and assurance status
We publish the state of each assurance item, not an aspiration. A planned milestone is not a completed assessment.
| Item | Status | Detail |
|---|---|---|
| SOC 2 Type I | Not started | A readiness program and control map exist internally; no auditor is engaged and no examination date has been set. The planned hosted cell is gated on Type I before it is offered. |
| SOC 2 Type II | Not started | Requires a 3–12 month observation window after controls operate; not credible before that window completes. |
| ISO 27001 | Not pursued | Not currently on the roadmap; SOC 2 is the first framework. |
| Penetration test | Not yet performed | A documented scope, test plan and disclosure policy exist. The engagement has not been executed and no report exists; this page will name the firm and the date once it has. |
| Internal security review | Ongoing | AI-assisted review supplemented by manual verification, with findings tracked and retested before release. An internal engineering control, not an independent audit. |
| Release integrity | In place | Release artifacts carry Ed25519 signature envelopes, SHA-256 sums and SPDX/CycloneDX SBOMs; verification works offline. Signing proves origin and integrity, not the safety of the code. |
| Cyber insurance | Not yet bound | Procurement follows incorporation of the operating entity, which is pending. |
3. Security practices
The detailed statement (design principles, assurance claims with their boundaries, and known limitations) is on the security page. The short version:
- Fail closedIf the decision journal cannot be written, the gateway denies the request and no credential is minted.
- Sealed recordsAppend-only, hash-chained journal with signed Merkle tree heads; clients verify exports offline with AAES out of the trust path.
- EncryptionWhere AAES holds credential material at all (the lab-only inline custody model), it is AES-256-GCM at rest with per-tenant keys; the production custody path keeps secrets in the client's own vault. TLS served in process, with plain HTTP refused unless explicitly opted in.
- Access controlCallers bound to their identity; agents cannot approve; irreversible verbs keep a human floor; operator SSO and MFA exist when configured.
- Honest limitsKnown limitations are published on the security page rather than left for you to find.
You can check the record claims yourself: the offline verifier is open source under Apache-2.0, and the sample verification pack includes a tamper demonstration.
4. Data handling and privacy
Client product data stays with the client. AAES is client-operated: records, identities, approvals and credentials live in the client's deployment. AAES receives no product traffic, and the daemon hashes action payloads rather than storing them. Optional support transfers are agreed in writing before anything is sent.
- Privacy policyWhat the website collects, legal bases and rights, retention, and transfer safeguards for correspondence.
- SubprocessorsProduct subprocessors: none engaged. Website and correspondence providers listed with roles, locations and safeguards.
- Data Processing AgreementA DPA covering the client-hosted product is drafted and under internal review. It is not yet available externally: it will be shared once counsel has reviewed it and the operating entity, whose incorporation is pending, can execute it, and will be published here after that.
- Data-subject requestsFor data we hold (website and correspondence), email hello@aaes.ai; we respond within the period required by applicable law, generally one month under the GDPR or UK GDPR, with any lawful extension explained. For data in a client's deployment, contact the client operating it.
5. Documents and requests
What we can share today, and on what terms:
When a SOC 2 report or a penetration-test report exists, its status and a letter of engagement or attestation will appear on this page. Until then, nothing on this page should be read as an independent assessment.
6. Report a vulnerability
Follow the vulnerability disclosure policy, or email hello@aaes.ai with the subject line Security. The same contact and policy are published as an RFC 9116 security.txt at /.well-known/security.txt. We acknowledge reports within five business days and provide a substantive assessment within fifteen.