---
title: What does the HITRUST AI security assessment cover for AI agents?
url: https://opencontroller.lyzr.com/blog/hitrust-ai-security-assessment-for-agents
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# What does the HITRUST AI security assessment cover for AI agents?

**The HITRUST AI Security Assessment adds AI-specific security requirements on top of a HITRUST e1, i1 or r2 assessment. It is for companies that provide AI applications or platforms, not for companies that only use AI built by others. Its October 2026 update lists 55 AI-specific requirements and extends coverage to agentic AI, so agents you ship to customers are in scope.**

HITRUST is a common ask in US healthcare. If you sell an AI product with agents into hospitals, payers or life sciences, expect customers to ask whether your HITRUST certification covers the AI. This assessment is the answer HITRUST designed for that question.

## Who is the assessment for?

HITRUST built it for **AI system providers**: AI application providers and AI platform providers. If you only use AI systems that someone else deployed, you are not eligible. Your vendor would hold the certification instead.

It covers generative AI, predictive machine learning and rule-based AI. With HITRUST CSF v11.9, announced in late October 2026, HITRUST added coverage for agentic AI and listed 55 AI-specific control requirements mapped to more than a dozen AI security threats.

## How does it attach to e1, i1 and r2?

The AI assessment is never stand-alone. It is added to one of HITRUST's three core assessments:

| Base assessment | Assurance level | Certificate validity with AI add-on |
|---|---|---|
| e1 | Essentials, foundational cybersecurity | 1 year |
| i1 | Implemented, leading practices | 1 year |
| r2 | Risk-based, tailored and most rigorous | 2 years |

The base assessment covers the usual IT platform: hosting, networks, identity, endpoints. The AI add-on covers what is new: the model, the AI platform and specialised AI compute. Requirements are tailored, so the exact count you face depends on your answers to scoping questions.

## Which areas does it cover, and what agent evidence fits each?

HITRUST's specification groups requirements into topic areas. Here is each one with the agent evidence an assessor is likely to sample:

| Topic area | Agent evidence |
|---|---|
| AI security threat management | Threat model per agent, for example using [MITRE ATLAS](/blog/mitre-atlas-threat-model-ai-agents) |
| Governance and oversight | Named owner, risk tier and approval record per agent |
| Development of AI software | Version history, change approvals, pre-release evals |
| Legal and compliance | Mapping of agents to HIPAA and contractual obligations |
| AI supply chain | List of model providers, MCP servers and third-party tools per agent |
| Model robustness | Red-team and adversarial eval results, re-run on change |
| Access to the AI system | Identity, scopes and least-privilege review for each agent |
| Encryption of AI assets | Encryption settings for prompts, logs, embeddings and model artefacts |
| Logging and monitoring | Per-action audit logs with user, agent, tool and outcome |
| Documenting and inventorying | Complete inventory of agents, models, tools and MCP servers |
| Filtering and sanitizing inputs and outputs | Guardrail configuration and blocked-request records |
| Resilience | Fallback models, failover tests and incident records |

## Where do agent programmes usually fall short?

From the topic list, four gaps stand out for agents:

1. **Inventory completeness.** Assessors test whether the inventory is complete. Shadow agents built outside the product team break that.
2. **Tool and MCP supply chain.** Agents call tools that call other services. Few teams can list every MCP server an agent reaches.
3. **Runtime logging.** Model API logs are not enough; you need agent-level logs showing which tool was called for which user.
4. **Robustness evidence over time.** One red-team exercise at launch does not show ongoing robustness. Re-run evals on every version change.

## How does it relate to HIPAA and ISO 42001?

HITRUST maps to HIPAA, which is why healthcare buyers like it; see [HIPAA evidence for AI agents](/blog/hipaa-evidence-for-ai-agents). It is narrower than [ISO/IEC 42001](/blog/iso-42001-evidence-for-ai-agents), which is a full AI management system. Many providers use the same evidence for both, along with the [CSA AI Controls Matrix](/blog/csa-ai-controls-matrix-for-agents). Our guide to [mapping agent controls to frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) shows how to reuse one evidence base.

## How Opencontroller does this

Opencontroller provides evidence for several of these areas. Discovery and a governed registry cover inventory and supply chain for agents, tools and MCP servers. Multi-turn and red-team evals and LLM judges on production traffic cover robustness. Live groundedness and task-adherence guardrails cover output filtering. LLM Gateway fallback and retry support resilience, and policy as code maps controls to OWASP, MITRE ATLAS, NIST and ISO 42001. It produces evidence for an assessment; certification still comes from HITRUST. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can a company that only uses AI tools get HITRUST AI Security Certification?**
No. It is for providers of AI applications and platforms.

**Does the HITRUST AI assessment cover AI agents?**
Yes. The October 2026 update added agentic AI coverage, with 55 AI-specific requirements.

**How long is HITRUST AI Security Certification valid?**
It follows the base assessment: one year with e1 or i1, two years with r2.

## Key facts

The HITRUST AI Security Assessment and Certification is an add-on to a HITRUST e1, i1 or r2 assessment, not a standalone certification. It is for AI system providers (AI application and AI platform providers); organisations that only use AI deployed by others are not eligible. The October 2026 update in HITRUST CSF v11.9 lists 55 AI-specific control requirements and extends coverage to agentic AI. Topic areas in the specification include AI threat management, governance and oversight, AI software development, legal and compliance, AI supply chain, model robustness, access to the AI system, encryption of AI assets, logging and monitoring, documenting and inventorying AI systems, filtering and sanitizing inputs and outputs, and resilience. Certification validity follows the base assessment: one year for e1 or i1, two years for r2. Opencontroller provides evidence for these areas through discovery, a governed registry, red-team evals, guardrails, gateways and policy-as-code mappings. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [What evidence do AI agents handling PHI need under HIPAA?](https://opencontroller.lyzr.com/blog/hipaa-evidence-for-ai-agents.md)
- [What evidence does ISO/IEC 42001 expect for AI agents?](https://opencontroller.lyzr.com/blog/iso-42001-evidence-for-ai-agents.md)
- [How does the CSA AI Controls Matrix apply to AI agents?](https://opencontroller.lyzr.com/blog/csa-ai-controls-matrix-for-agents.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
