NIST’s AI Risk Management Framework is a voluntary, cross-sector framework for managing AI risk.
National Institute of Standards and Technology · verified 2026-09-22Industry / Technology
AI security assessment for Technology
Secure the AI you ship and the AI your teams use to build, operate, and support it.
01 / CONTEXT
Technology companies have to secure both sides of AI adoption.
Technology organizations face two AI-security surfaces at once: AI embedded in products and AI adopted internally by engineering, product, support, and operations teams. Both can create new paths to sensitive data, privileged tools, software supply chains, and customer-facing decisions.
NIST’s AI Risk Management Framework is explicitly cross-sector and intended to help organizations incorporate trustworthiness considerations into the design, development, use, and evaluation of AI systems. NIST’s Generative AI Profile adds actions for risks specific to generative AI, while its Cyber AI work continues to develop cybersecurity-focused guidance.
Ayliea’s technology assessment maps the actual system rather than assuming one framework answers every question: model and API inventory, data flows, agent permissions, SDLC use, third-party dependencies, prompt-injection exposure, logging, incident response, and the evidence available to support customer or security review.
02 / Current signals
What shapes the review.
Primary-source requirements and guidance establish the context. The engagement still follows the actual system, data, and use case in scope.
NIST AI 600-1 identifies risks and risk-management actions specific to generative AI.
NIST Generative AI Profile · verified 2026-09-22NIST is developing a Cybersecurity Framework profile focused on cybersecurity risks associated with AI systems and capabilities.
NIST Cyber AI Profile project · verified 2026-09-2203 / Requirements & frameworks
Map evidence to the environment around the AI.
Frameworks are not treated as interchangeable certifications. Ayliea uses the requirements and guidance that actually belong in the engagement scope.
NIST AI RMF
A voluntary cross-sector framework for incorporating trustworthiness considerations into the design, development, use, and evaluation of AI systems.
NIST Generative AI Profile
NIST AI 600-1 identifies generative-AI risks and suggested risk-management actions that can inform product and internal-use reviews.
NIST Cyber AI work
NIST is developing a Cybersecurity Framework profile addressing cybersecurity risks associated with AI systems and AI-enabled security capabilities.
Customer & assurance requirements
SOC 2, ISO/IEC 27001, customer contracts, privacy obligations, and sector-specific requirements may shape evidence expectations depending on the company and product in scope.
04 / Assessment surface
What we examine in Technology.
The exact procedures are scoped to the organization. These are common areas that shape evidence requests and assessor judgment.
AI Product Attack Surface
Map models, prompts, retrieval components, agents, tools, APIs, plugins, and data stores that form the AI-enabled product surface, including trust boundaries and externally reachable paths.
Prompt Injection & Agent Permissions
Assess how untrusted input can influence models or agents, what tools and data those systems can reach, and whether authorization, isolation, confirmation, and least-privilege controls limit blast radius.
AI in the SDLC
Review code assistants, autonomous development agents, CI/CD integrations, repositories, secrets, and generated code practices for data leakage, insecure output, excessive permissions, and software-supply-chain risk.
Customer Data & Tenant Isolation
Trace customer information through AI features and providers, reviewing retention, training/reuse, tenant boundaries, access controls, observability, deletion, and alignment with customer commitments.
Model, API & Vendor Supply Chain
Inventory hosted models, AI APIs, open-source components, embeddings, vector stores, and other dependencies; evaluate provenance, update paths, security evidence, incident obligations, and concentration risk.
AI Detection & Incident Response
Determine whether AI-specific abuse, anomalous agent behavior, model or prompt attacks, credential misuse, and provider incidents can be logged, detected, investigated, contained, and communicated to affected customers.
05 / Common review needs
Start with the problem you are actually trying to solve.
These focused pages go deeper on common security and diligence questions in this industry.
01
AI Agent Security Assessment
Assess prompt-injection exposure, agent identity and authority, tool permissions, sensitive-data access, external actions, and the controls limiting an agent's blast radius.
Explore this review →02Prompt Injection Security Assessment
Evaluate direct and indirect prompt-injection paths across LLM applications, retrieval pipelines, agents, tools, and connected data sources.
Explore this review →06 / OUTPUT
Evidence a reviewer can interrogate.
Ayliea documents the systems reviewed, evidence examined, findings, limitations, framework mappings, remediation priorities, and accountable human review. The report is designed to support diligence and internal risk decisions—not to substitute for legal advice or a regulator’s determination.
ASSESSMENT RECORD
AI system inventory
Data-flow & boundary review
Evidence references
Findings & limitations
Industry-relevant mappings
Remediation priorities
Accountable sign-off
07 / SCOPE THE WORK
Start with the AI system and the evidence you already have.
We will determine whether an independent assessment fits the review, customer, compliance, or security question you need to answer.