The system of record for AI activity
A tamper-evident record of what your AI system did.
For AI products with enterprise customers whose security reviews require an audit trail.
A signed audit chain that third parties verify without an Aleutian account. End-user DSAR portal for GDPR Articles 15 and 17. Two-day API integration.
The ledger view. Lifecycle state, platform health, and the signed trust manifest, shown with representative data.
The security review requirement.
Enterprise security reviews now routinely require an AI audit trail. Most AI products do not have one. Building one in-house costs an engineering quarter on infrastructure that ships no customer-facing feature, and the deal waits on the result.
The in-house build
Components required to satisfy the review.
- Tamper-evident hash chain over every prompt and response
- Key rotation, deletion certificates, regulator-acceptable canonicalization
- Signed evidence bundles for SOC 2, GDPR, HIPAA, NIST 800-53
- An offline verifier their CISO can actually run themselves
- 4–6 months for v1, 6–12 more on edge cases
If your product is a high-risk AI system under the EU AI Act, Articles 12 and 19 also apply. The requirements are covered in the 2026 section below.
The life of your evidence.
The dashboard is organized around the lifecycle of a single piece of evidence. Six stages take an event from ingestion to offline verification. Every state shown is signed, and the chain panel exposes each entry's hash and its reference to the previous entry.
The chain panel, shown with representative data. Every entry references the previous entry's hash. A change anywhere in history breaks every hash after it.
01 · INGEST
Ingest
Your product pushes prompts, responses, and consent actions through the SDK. Each push is a signed chain entry attributed to a source and API key.
02 · LINK
Link
Entries are linked into the hash chain on a five-minute cadence. Each entry references the hash of the previous entry.
03 · VERIFY
Verify
A scheduled verifier re-walks the chain and reports breaks. The ledger shows the lifetime break count.
04 · ANCHOR
Anchor
Chain heads are sealed to your storage bucket as signed anchors. Anchors are the reference points third parties verify against.
05 · EXPORT
Export
Admin-approved JSONL exports to your bucket, with a required compliance framework mapping. Chain data stays sealed to your registered keys.
06 · VERIFY-EXPORT
Verify offline
Exported bundles verify with the open-source CLI, your key, and no dependency on Aleutian infrastructure.
Independent verification.
verify.aleutian.ai
A verify link for any sealed record lets a security team confirm the signature in their own browser without an Aleutian account. The full trail is verified offline against your exported bundle with the open-source SDK.
The receipt your data subject takes with them.
When your end user requests their data, the receipt they leave with is a signed list of every processing event you recorded about them — timestamps, signing keys, chain anchors. Nothing your team had to assemble by hand.
With that list, the data subject exercises the full Article 15 right of access by requesting decryption of the underlying payloads under Article 15(3). You hold the encryption key and fulfill the decryption.
Aleutian operates only as the cryptographic anchor that proves the events happened, who recorded them, when, and under which consent. Your data subjects, your regulators, and your own team get a record where any rewrite is detectable and verifiable against signed anchors your customers hold.
The request queue, shown with representative data. Access and erasure requests tracked against the 30-day SLA, with a signed evidence pack for the auditor.
Who Aleutian is for.
You're a CTO or founder at an AI agent platform or AI-native product with paying enterprise customers.
The blocking issue is the audit-trail requirement in enterprise security review. Aleutian is the buy-versus-build answer to that requirement. The same record layer is the accountability infrastructure that makes agents acceptable to ship in production.
Aleutian fits when
- An enterprise prospect's security review has asked you for an AI audit trail and you don't have one
- The in-house alternative is a quarter of engineering on infrastructure that ships no customer-facing feature
- You need evidence to provide to a security reviewer this week rather than in six months
- You have enterprise prospects in security review (design partners are the exception)
- Your AI product runs on one or more major model providers (Vertex AI, Anthropic, OpenAI, or any combination)
Three requirements converge on the same record layer. EU AI Act logging obligations, GDPR access, erasure, and portability rights, and post-quantum signature requirements in federal procurement. Aleutian produces signed, independently verifiable evidence for all three today.
Different shape of company? Jump to who else we work with →
Founded by the engineer who managed data engineering for Apple Health, working between the data and privacy engineering teams.
Two days of integration. Three steps.
Embed the SDK.
Your AI product calls the Aleutian API on every prompt, response, and consent action via our Go, Python, or JavaScript libraries. The integration takes around two days, including wiring up your existing identity layer.
We sign and chain every event.
Each entry is hashed with SHA-512, signed with ML-DSA-65 (FIPS 204, post-quantum) using a per-tenant Cloud KMS key, and chained into an append-only, cryptographically ordered log. The default operating mode is zero-knowledge — content forwards to your AI provider and is never stored on our side. Encrypted-storage and SDK-only modes are available for stronger guarantees.
Your customers and their regulators verify it themselves.
Open-source verifier SDKs let any third party — your enterprise prospect's CISO, an auditor, or a regulator — verify your chain offline against our published public keys at verify.aleutian.ai, with no dependency on Aleutian.
For the full visibility-modes breakdown (zero-knowledge default, encrypted-and-chained, consented PII scan, SDK-only), see the compliance page →
What Aleutian isn't.
We're not a privacy platform. We don't replace OneTrust, Transcend, or DataGrail — your privacy team keeps their workflow, dashboard, and training; their existing tooling calls our API for the cryptographic layer underneath.
We're not an AI observability tool. Helicone, Portkey, and LangSmith tell you what your AI did so you can debug it. We tell a regulator, court, or auditor what your AI did so it holds up as evidence.
We're not commodity logging. The right comparison for our pricing is your AI inference bill (typically 1–3%), not your Datadog bill.
Why this matters in 2026.
The EU AI Act logging mandate.
The EU AI Act's high-risk obligations have been in force since August 2, 2026. Article 12 requires automatic logging of events over a system's lifetime; Article 19 requires providers to retain those logs for at least six months. Penalties reach €15M or 3% of worldwide annual turnover. Aleutian's audit chain produces logs that are tamper-evident and independently verifiable, which is exactly what those articles require you to be able to show.
The post-quantum transition.
NIST published FIPS 204 in August 2024, and the CNSA 2.0 timeline targets 2030–2035 for full migration across federal systems. Over the next eighteen months, enterprise procurement reviews will increasingly ask AI vendors about their post-quantum cryptography roadmap, and Aleutian's signing layer already operates on ML-DSA-65 today rather than as a planned future upgrade.
Your agent will fail before an attacker does.
Documented incidents include agents deleting production data and fabricating records of the deletion afterward. When the system that caused the failure also produces the logs, an external tamper-evident record is the only reliable account of what happened.
The audit-chain expectation.
Enterprise security reviews now routinely ask for a tamper-evident record of AI usage. The vendors who can hand back a signed chain win deals; the ones who hand back screenshots of dashboards don't.
Different shape of company?
You're a DPO at an AI-native company with EU users. GDPR Articles 15, 17, and 20 already apply, and the EU AI Act logging obligations are in force. Aleutian's DSAR pipeline and signed deletion certificates sit underneath OneTrust / Transcend / your in-house tooling. Read the GDPR map →
You're a state agency, university system, or large K-12 district. Aleutian deploys on Google Cloud with per-tenant Service Account isolation auditable in your own Cloud Audit Logs. StateRAMP scoping under way; we're in active conversations with the Google Public Sector partner program. Read the public-sector map →
You're a federal agency or DoD contractor. Post-quantum from day one (FIPS 204 / ML-DSA-65), FedRAMP authorization on the roadmap with Google sponsorship in discussion, SBIR-eligible. Read the NIST + federal map →
You're an AI infrastructure provider thinking about embedded compliance. Your platform's findings, evals, or traces land on our chain, signed with your key — the evidence layer inside your product. Design partners only at this stage. Book a call →
Take a 20-minute call.
We cover what enterprise security reviews require, what an integration looks like, and whether Aleutian is the right answer. If the fit is not there, we say so.