AI trust, risk and security management (AI TRiSM) is a framework coined by the research and advisory firm Gartner for the set of practices and technologies an organization uses to keep its AI systems trustworthy, compliant, reliable and secure. It groups oversight, data protection, monitoring and security controls under one name so they can be planned together. It is an analyst framework and a market label, not a published standard or certification.
At a glance
- AI TRiSM is Gartner’s term, not a standards-body specification, and Gartner has refined how it groups the framework over time.
- It brings together areas such as AI governance and inventory, runtime monitoring and policy enforcement, information protection, and security of the underlying AI infrastructure.
- It applies to AI the organization builds and to third-party AI tools, apps and agents it uses.
- Many vendors market products as AI TRiSM tools; most typically cover only part of it.
- Published frameworks such as the NIST AI Risk Management Framework are often used alongside it.
What problem it solves
AI risk tends to be split between teams. Legal looks at regulation, security looks at attacks such as prompt injection, data teams look at privacy and data quality, and business owners look at accuracy. Each may buy its own tools or set its own rules, leaving gaps, especially for shadow AI that nobody approved.
AI TRiSM gives these concerns one frame. It treats AI risk as something to manage continuously across the life of each AI system, from selection and testing through daily use, not only at approval time. For a buyer, it is mainly a checklist for spotting gaps and a vocabulary for understanding vendor pitches.
How it works
Gartner describes AI TRiSM in layers or pillars that it has updated as the market has changed. In plain terms, the areas usually include:
- Governance and inventory. Knowing which AI models, applications and agents are in use, who owns them and what policies apply. This is where AI governance sits.
- Runtime inspection and enforcement. Monitoring AI inputs, outputs and actions while systems run, and applying AI guardrails to block or flag policy violations, unsafe content or anomalies.
- Information protection. Controlling what data AI systems can see and share, through classification, access controls and data loss prevention.
- Infrastructure and model security. Protecting the models, pipelines and hosting environments, along with traditional security controls.
Supporting practices include explainability, model monitoring for drift and accuracy, privacy protection and human-in-the-loop (HITL) review. Organizations usually build this from a mix of existing security, governance, risk and compliance (GRC) and data protection tools plus newer AI-specific products.
When it matters for buyers
- When AI use spreads beyond pilots. Once many teams use many AI tools, a framework helps you see the whole picture.
- When evaluating AI security products. Knowing the layers helps you see which part of the problem a product covers and what you would still need.
- When the board or customers ask about AI risk. It offers a structured way to report on controls, not just policy.
- When deploying agents. Agents that act on systems raise the stakes for runtime monitoring and enforcement. See our governance, risk and compliance overview for help assessing tools.
Questions to ask vendors
- Which parts of AI TRiSM does your product address, and which does it leave to other tools?
- Can you discover AI models, apps and agents in use, including ones we did not approve?
- Do you inspect AI inputs and outputs at runtime, and what can you block versus only alert on?
- How do you integrate with our existing data protection, identity and security tools?
- Which published frameworks, such as the NIST AI RMF or ISO/IEC 42001, can your reporting map to?
- Which AI platforms and models do you support today, and how quickly do you add new ones?
How it differs from the NIST AI RMF
The NIST AI Risk Management Framework is a voluntary framework published by a US government standards agency. It describes outcomes and activities for managing AI risk and is free to use and cite. AI TRiSM is an analyst framework from a private research firm, and it leans more toward the technology categories and controls used to carry out AI risk management, which is why vendors use it to position products. The two are compatible: many organizations use the NIST AI RMF, or ISO/IEC 42001, as the reference they align with and report against, and use AI TRiSM’s layers as a way to check which technical controls they still lack.