> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ayliea.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Security Assessment (AISS)

> Score your AI surface against the open Ayliea AI Security Standard — 10 control domains, 56 sub-controls, 9 framework crosswalks, published under CC-BY-4.0.

# AI Security Assessment (AISS)

AISS — the **Ayliea AI Security Standard** — is the open methodology behind every AI Security assessment on the platform. The full specification is published at [github.com/Ayliea/aiss](https://github.com/Ayliea/aiss) under the [Creative Commons Attribution 4.0 license](https://creativecommons.org/licenses/by/4.0/), so you can verify any score from the spec, fork the standard for your own internal use, or propose changes through the public RFC process.

The current production version is **v1.2.3**: 10 control domains, 56 sub-controls, and 9 framework crosswalks to NIST CSF 2.0, ISO/IEC 27001:2022, NIST AI RMF, NIST AI 600-1, CIS Controls v8.1, EU AI Act, Colorado AI Act, OWASP LLM Top 10, and MITRE ATLAS.

<Note>
  AISS is the only framework available on the Free tier, so any organization can run their first assessment without payment or sales contact.
</Note>

## Why an open standard

Most AI security scoring is opaque — proprietary algorithms with no way for an auditor to reproduce the math. AISS takes the opposite position: every score is fully derivable from the answers you provide and the published spec. Anyone with the standard and your answers can recompute your score line-by-line.

That transparency matters in three places:

* **In the audit room** — hand auditors the published JSON spec; they verify framework crosswalks, scoring rubric, and sub-control citations against their own reference frameworks.
* **At the boardroom** — board-ready reports cite a standard your board's outside counsel can read directly.
* **In your security program** — fork AISS internally, extend it for your environment, or propose upstream changes through the public RFC process.

## The 10 control domains

The framework spans 87 questions across 10 AI-specific control domains. Each domain maps to one or more clauses in the source frameworks above.

<AccordionGroup>
  <Accordion title="AC-1 — AI Governance & Policy">
    Organizational policies, roles, and accountability structures for AI adoption and oversight. Includes acceptable-use policies, executive AI literacy, AI ethics committees, and the connection between AI use and the broader risk-management program.
  </Accordion>

  <Accordion title="AC-2 — AI Asset Management">
    Discovery, inventory, classification, and lifecycle management for every AI tool, model, agent, and integration in your environment. Covers AI development tool governance (Copilot, Cursor, etc.) and shadow-AI detection.
  </Accordion>

  <Accordion title="AC-3 — Data Protection in AI">
    Controls for data flowing to, from, and within AI systems. Classification, encryption, minimization, retention, RAG and vector-store security, and synthetic-content provenance.
  </Accordion>

  <Accordion title="AC-4 — Access Control for AI">
    Authentication, authorization, privileged-access management, and least-privilege access to AI tools and services. Covers human-to-AI, AI-to-system, and agent-to-agent authorization.
  </Accordion>

  <Accordion title="AC-5 — AI Supply Chain Security">
    Vendor assessment, contract review, model provenance, and continuity planning for third-party AI services. Includes assessing model origin, training-data lineage, and sub-processor exposure.
  </Accordion>

  <Accordion title="AC-6 — AI Output Validation">
    Human oversight, accuracy monitoring, bias detection, content disclosure, automated quality controls, and synthetic-content marking (per EU AI Act Article 50 and similar deepfake-disclosure laws).
  </Accordion>

  <Accordion title="AC-7 — AI Incident Response">
    Incident planning, reporting, classification, containment, post-incident review, and regulatory notification for AI-specific events — including prompt-injection attacks, data leakage via AI output, model compromise, and agent guardrail failures.
  </Accordion>

  <Accordion title="AC-8 — AI Monitoring and Logging">
    Usage logging, anomaly detection, prompt/response logging, performance monitoring, and log integrity for AI systems. Connects AI activity to your existing SIEM and incident-response workflows.
  </Accordion>

  <Accordion title="AC-9 — AI Training & Awareness">
    Security awareness programs, role-based training, acceptable-use communication, threat awareness (phishing-via-AI, social engineering of AI agents), and executive AI literacy.
  </Accordion>

  <Accordion title="AC-10 — Model Security">
    Model access controls, adversarial testing, versioning, integrity verification, prompt-injection defense, and agentic AI guardrails (per MITRE ATLAS T0080–T0112 and OWASP LLM06 Excessive Agency).
  </Accordion>
</AccordionGroup>

## Glass-Box Score breakdown

For every AISS assessment, the results page includes a per-domain expandable breakdown. Each AC-N panel shows the questions answered, weight per question, points earned vs. maximum, MITRE ATLAS technique mappings, framework crosswalks, and a deep-link to the matching domain markdown in the public [Ayliea/aiss](https://github.com/Ayliea/aiss) repository. Any auditor can hand-calculate the score from the published spec plus your answers — no proprietary algorithm involved.

A separate **AISS Coverage** page at `/aiss` rolls all 10 domains into a single posture grid so you can see your overall AI security posture in one view, with each domain card linking directly to the corresponding spec page.

## Propose a change

If a question or scoring rubric doesn't match your environment's reality, every question row in the Glass-Box drilldown has a pencil icon that opens a pre-filled GitHub issue against [Ayliea/aiss](https://github.com/Ayliea/aiss). The RFC issue template captures the control ID, question text, your selected option, and points earned automatically — practitioners shape the standard through the same public process they use to consume it.