Short answer
AIUC-1 is an agent-scoped third-party attestation. Six control domains: data & privacy, security, safety, reliability, accountability, society. Ships on a quarterly release cycle with annual customer recertification. It sits on top of SOC 2 (baseline) and ISO 42001 (organisation-wide AI programme) — not instead of them.
The 6 domains
Data & privacy
What customer and third-party data crosses into the agent's context window. Redaction, retention, deletion on request, vendor DPAs for LLM providers.
Security
Agent authentication and identity, tool-scope allow-listing, output logging, prompt-injection and adversarial-input resistance.
Safety
Action authorisation and human-in-the-loop thresholds for high-impact operations. Red-team findings and mitigations.
Reliability
Drift and quality monitors, incident response for runaway loops or cost spikes, kill-switch procedure, evaluation on model updates.
Accountability
Ownership, audit trails for every tool invocation, incident post-mortems, affected-party redress.
Society
Impact assessment for the agent's intended use, affected groups, foreseeable harms, benefit/harm trade-off.
Domain names reflect common practitioner shorthand in 2026. For the authoritative control list and current version numbering, see the AIUC-1 site.
Quarterly cycle, annual certification
AIUC-1 ships a new version each quarter, with changes tied to a customer's certification cycle. In practice:
- Each customer is pinned to a quarterly version at the time of their certification; that version applies through the next recertification.
- Platform support follows: Drata introduced native AIUC-1 support on 16 July 2026 and shipped the July 2026 version update on 2 October 2026 (source).
- No big-bang breakage: because versions are pinned per customer, a quarterly change doesn't invalidate existing reports. You roll onto the newer version at your next assessment.
Readiness checklist
Decide agent scope
Pick one AI-agent product (not the whole company). Document the agent's intended use, the tools it can invoke, and the data it touches.
Stand up the six-domain control pack
Policies, procedures and evidence for each of the six domains. Reuse SOC 2 CC6/CC7/CC8 and ISO 42001 A.6/A.9 where they apply; add agent-specific controls (tool-scope allow-list, action approval threshold) on top.
Run adversarial testing
A red-team pass covering prompt injection, jailbreak, data exfiltration via tool chaining, cost spikes. Log every (prompt, response, severity, category) tuple as evidence.
Pick an assessor
Verify the firm on the AIUC assessor directory. Request quotes from 2–3 qualified assessors. For analogous agent-scoped engagements, 4–12 weeks is typical.
Ship the report to the trust centre
Publish the AIUC-1 report alongside your SOC 2 and ISO 42001 artefacts. Set a calendar reminder for the next quarterly re-certification.
How AIUC-1 relates to SOC 2 and ISO 42001
| Dimension | AIUC-1 | SOC 2 | ISO 42001 |
|---|---|---|---|
| Scope | One AI-agent product | Service organisation | Organisation-wide AI programme |
| Attestation style | Agent-specific third-party report | AICPA attestation | Management-system certification |
| Cycle | Quarterly releases, annual recertification | Annual (Type 2 observation) | Three-year certificate, annual surveillance |
| What it proves | The agent's action surface is bounded, logged and overseen | Information-security controls work | AI governance across the organisation is operational |
For a decision tree on which to pursue first, see AIUC-1 vs SOC 2 vs ISO 42001 and What is AIUC-1?
AIUC-1 readiness with LowerPlane
LowerPlane maps the six AIUC-1 domains to the same evidence base as SOC 2 and ISO 42001. Controls you build for either framework also prepare you for an AIUC-1 third-party assessment. Native AIUC-1 reporting is on our roadmap; current support is readiness with mapped controls.
Who audits AIUC-1
Third-party assessors sign AIUC-1 reports. Some firms publicly state they operate as AIUC-1 auditors; the authoritative list is the AIUC assessor directory. Before accepting a report, buyers should check the signing firm is listed there and ask about the firm's AIUC-specific qualification process.
Need an auditor?
Our sister site auditxyz.com matches AI startups to firms running AIUC-1, SOC 2 and ISO 42001 engagements. (Related-site disclosure: auditxyz.com is operated by the LowerPlane team.) Verify any firm against the AIUC assessor directory before contracting.
Frequently asked questions
Is AIUC-1 a replacement for SOC 2?
No. SOC 2 is an AICPA attestation covering the service organisation's security controls. AIUC-1 is an agent-specific attestation covering one AI-agent product. Enterprise buyers typically ask for both: SOC 2 as the baseline, AIUC-1 as the agent-specific overlay. See our AIUC-1 vs SOC 2 vs ISO 42001 comparison for the full matrix.
How often does AIUC-1 change?
AIUC-1 ships on a quarterly release cycle. Each version is tied to a customer certification period, so a report issued under one quarterly version is valid until the next re-certification. Compliance platforms (Drata introduced native AIUC-1 support on 16 July 2026 and shipped the July 2026 version update on 2 October 2026) track these versions and alert customers when controls shift.
Does LowerPlane support AIUC-1?
LowerPlane currently provides AIUC-1 readiness with mapped controls — the six domains (data & privacy, security, safety, reliability, accountability, society) are tracked against the same evidence base as SOC 2 and ISO 42001, so the controls you build for either framework also prepare you for an AIUC-1 third-party assessment. Native AIUC-1 reporting is on the roadmap; book a demo for current status.
Who signs an AIUC-1 report?
An assessor firm. The right verification step is to check the AIUC assessor directory for the firm named on the report. Some compliance platforms (e.g. A-LIGN) publicly state they are AIUC-1 auditors; verify against the AIUC site before contracting.
Should a pre-SOC 2 startup pursue AIUC-1?
Usually no. AIUC-1 is an agent-specific overlay that sits on top of an existing security attestation. If you don't have SOC 2 Type 2, finish that first — procurement teams reading an AIUC-1 report on its own will still ask for SOC 2.
When does AIUC-1 matter for an AI startup?
When you ship an AI agent (a system that takes autonomous actions via tool use, browsing, code execution), and enterprise prospects ask about agent-specific controls by name. If your product is read-only or research-only, SOC 2 and ISO 42001 usually cover the procurement ask without AIUC-1.
Get AIUC-1 ready
Book a 20-minute call. We'll scope which AIUC-1 domains apply to your agent and map them to the SOC 2 and ISO 42001 work you already have.
Book a demo