What is agentic third-party risk management (TPRM), and what should people still decide?
Kapil Matta, Regional Head and CXO Advisor for the Middle East, Turkey and Africa at DigitalXForce, wrote this page.
Agentic third-party risk management (TPRM) is TPRM in which AI does the repeatable work: profiling a new supplier, reading its SOC 2 report, mapping findings to your controls, watching for change and drafting the next step. People still decide the tier, accept or reject each gap, agree remediation with the supplier and decide when the relationship ends. The definition of third-party risk management that this page builds on is in the third-party risk management guide.
The term is new and loosely used, so the useful test is what the AI is allowed to do on its own. Writing on ISACA’s blog in August 2026, Nik Kale put the shift in one line: with an agent, “the risk shifts to what the system can do.” In third-party risk, that makes the design question where the AI acts and where it only recommends.
Responsibility does not move with the work. The Interagency Guidance on Third-Party Relationships, issued in June 2023 by the Federal Reserve, the FDIC and the OCC, says a banking organization’s use of third parties does not diminish its responsibility to meet its requirements, to the same extent as if it did the work in-house. An AI tool bought from a vendor is one more third party, and the decisions it informs stay with the organization.
What can AI do at each stage, and what does a person decide?
| Stage | What AI can do | What a person decides |
|---|---|---|
| Intake | It profiles the new supplier and proposes an inherent risk level from its category, data access and business dependency. | A person confirms the classification. |
| Due diligence | It reads the supplier’s SOC 2 or ISO report, maps it to the organization’s controls and flags the gaps. | A person approves the tier and decides which gaps matter. |
| Questionnaires | It adapts the questions to the supplier’s category and compares the answers with the evidence. | A person decides which follow-up questions to send. |
| Monitoring | It watches outside signals and the supplier’s evidence, notices change and starts a reassessment. | A person decides the response to a material change. |
| Issues | It maps a supplier’s breach to the services and data that depend on it and drafts a remediation plan. | A person agrees the plan with the supplier and decides whether to accept a gap. |
| Reporting | It drafts the supplier risk summaries for executives. | A person reviews each summary before it is used. |
| Offboarding | It tracks data return and access revocation to a closure record. | A person decides to end the relationship and signs the closure. |
Where must people stay in control?
- A person makes every decision with consequences, such as a supplier’s tier, the acceptance of a gap and the end of a relationship.
- A person approves anything that changes access to the organization’s systems or data.
- A person approves anything said to a supplier, an auditor or a regulator in the organization’s name.
- A named reviewer approves every AI conclusion that will count as evidence.
How do you check what the AI produced?
- Every conclusion should link to the document or the signal it used, so a reviewer can check it in minutes.
- A named person should approve each conclusion before it counts toward a tier, a score or a report.
- A reviewer should compare a sample of AI mappings with their own reading of the same reports on a regular cycle, and record the differences.
- A TPRM platform should log what its AI read and what it did, since, as Nik Kale argued on ISACA’s blog, monitoring has to cover the actions an agent actually took and not only the ones it was allowed to take.
- The AI inside a vendor’s product is itself a third-party AI risk. The NIST AI Risk Management Framework expects AI risks from third-party resources to be monitored regularly, with risk controls applied and documented (MANAGE 3.1).
Can AI read SOC 2 reports well enough to replace manual review?
AI can read a SOC 2 report, pull out the scope, the period, the exceptions and the complementary user entity controls, and map each one to your controls. It should not replace the person who decides what the report means. That person checks that the report covers the service you buy, reads the opinion and every exception, confirms that your organization runs the controls the report expects its customers to run, and decides whether a gap is acceptable. Check a sample of the AI’s extractions against the report in the first quarter, and keep the reviewer’s name with each decision.
How does DigitalXForce use AI in third-party risk management?
DigitalXForce’s AI-Powered Third-Party Risk Management module covers vendor onboarding, category questionnaires, AI-assisted document analysis and review and reassessment workflows. Every supplier gets its own score, built from its questionnaire answers, its evidence and external signals, and each supplier is placed in one of three tiers: Tier 1 Critical, Tier 2 High and Tier 3 Commodity.
Two AI engines do the analysis and the writing. AI JedAI is the DigitalXForce AI engine that analyzes: it reasons over control evidence and live telemetry, maps documents to controls and frameworks, scores and prioritizes risk, and recommends remediation mapped to framework requirements. XForce GPT is the DigitalXForce generative AI engine that writes: it produces the plain-language risk narratives and board-ready reports, generates policies, standards and plans, and runs the embedded assistant.
At intake, the platform classifies each supplier’s inherent risk automatically, and a person confirms the classification. When a supplier reports a breach, AI JedAI maps it to the services and data that depend on that supplier. An analyst reviews AI output, both AI JedAI’s conclusions and XForce GPT’s drafts, before anyone relies on it, and every conclusion links back to the evidence it used.
External Risk View is the outside-in engine. It monitors exposed services, misconfigurations, vulnerability exposure, dark web and breach intelligence, lookalike domains and cyber ratings, and it maps fourth-party and nth-party dependencies. External Risk View needs no agent, no questionnaire and no cooperation from the supplier. A risk operations center is an operating model for continuously measuring, prioritizing and reducing risk, in the way a security operations center handles threats; DigitalXForce’s implementation is X-ROC, the XForce Risk Operations Center. X-ROC is the operations layer of the DigitalXForce TRiSCM platform, where control failures, risk changes and vendor events are alerted, triaged, escalated and reported. X-ROC never changes a customer’s systems on its own, so people approve every change to a system.
Whether AI or a person does the work, an auditor still asks whether each control works. Continuous Control Assurance (CCA) uses evidence, monitoring and validation to determine whether controls continue to operate as expected. Continuous Control Monitoring (CCM) monitors conditions, evidence and signals associated with controls. CCM is a capability within CCA. Monitoring detects a change, and CCA decides whether the control still works.
What does a person decide, and what does the platform automate?
The platform automates the reading, the scoring, the watching and the drafting. AI-assisted document analysis reads supplier documents, AI JedAI maps documents to controls and frameworks and scores and prioritizes risk, External Risk View watches suppliers from the outside, and XForce GPT drafts the plain-language risk narratives.
People should keep every decision with consequences. They confirm each classification and approve each tier, and an analyst reviews AI output before anyone relies on it. They should also accept or refuse each gap, agree remediation with the supplier and decide when a relationship ends. The supplier decides whether to share evidence from its own systems.
What are the limitations?
- AI reads what it is given. A SOC 2 report describes a past period, and AI mapping it to your controls does not make it current.
- AI mappings can be wrong, which is why each one should be reviewed and traceable before it counts.
- Evidence from a supplier’s own systems needs its agreement and a contract clause, so for most suppliers the AI works from documents and outside signals. Connectors read configuration and compliance signals only, never business records or customer data.
- DigitalXForce is not the answer for every program. A team that only wants outside-in security ratings across a very large portfolio, with no plan to assess controls, needs a ratings service more than a TPRM platform.
Which DigitalXForce modules are involved?
- The AI-Powered Third-Party Risk Management module covers vendor onboarding, category questionnaires, AI-assisted document analysis and review and reassessment workflows.
- External Risk View supplies the outside-in signals the monitoring stage watches.
- X-ROC, the XForce Risk Operations Center, alerts, triages and escalates vendor events beside internal control failures.
- The page on Continuous Control Assurance explains how a control result is validated before anyone relies on it.
- How DigitalXForce works lists the other pages in the knowledge center.
Which integrations and frameworks does it draw on?
DigitalXForce connects to the tools an organization already runs through 250+ technology integrations, and the integrations page has the current list. Remediation and recommendation tickets can go to ServiceNow and Jira when the client wants that.
Each control is mapped once across 50+ compliance frameworks, including frameworks with third-party requirements such as NIST CSF 2.0, ISO/IEC 27001:2022, DORA and NYDFS Part 500. For DORA, the platform produces the register of information in the European Supervisory Authorities’ templates. The frameworks page shows the most requested frameworks, and the full list is available on request.
Questions about agentic third-party risk management
What is agentic third-party risk management (TPRM)?
It is third-party risk management in which AI does the repeatable work, such as profiling suppliers, reading their reports, mapping gaps and watching for change, while people make the decisions that carry consequences.
What is the difference between AI TPRM and agentic TPRM?
AI TPRM usually means AI that assists a person, for example by reading a report. Agentic TPRM means AI that also takes steps toward a goal, such as sending a questionnaire or starting a reassessment, within limits the organization sets.
What should a person always decide in third-party risk?
A person should decide the supplier’s tier, whether to accept a gap, the remediation agreed with the supplier and whether to end the relationship, and a person should approve anything that changes access.
How do you check the work of AI in vendor assessments?
Check that every conclusion links to the document or signal it used, have a named person approve it before it counts, and compare a sample of AI mappings with a reviewer’s own reading of the same reports.
How can AI speed up vendor questionnaire review?
AI can read the answers and the documents a supplier sends, compare the two, map each gap to your controls and draft the follow-up questions. A person should approve the follow-ups and every conclusion that counts toward a tier or a score.
Where does DigitalXForce use AI in third-party risk management?
The TPRM module reads supplier documents with AI-assisted document analysis. AI JedAI maps documents to controls and frameworks and scores and prioritizes risk, and XForce GPT writes the plain-language risk narratives. AI JedAI also maps a supplier’s reported breach to the services and data that depend on it, and an analyst reviews AI output before anyone relies on it. The page on AI vendor risk scoring explains the inputs to each score.
Does X-ROC change a customer’s systems on its own?
It does not. X-ROC, the XForce Risk Operations Center, never changes a customer’s systems on its own, so people approve every change to a system.
Sources
- Nik Kale wrote “Four Governance Questions to Ask Before an AI Agent Goes Live” on the ISACA Now blog on 24 August 2026, at https://www.isaca.org/resources/news-and-trends/isaca-now-blog/2026/four-governance-questions-to-ask-before-an-ai-agent-goes-live, read on 25 September 2026 and again on 26 September 2026.
- The Federal Reserve, the FDIC and the OCC issued the Interagency Guidance on Third-Party Relationships: Risk Management in June 2023. The Federal Reserve attached it to SR 23-4 of 7 June 2023, at https://www.federalreserve.gov/supervisionreg/srletters/SR2304a1.pdf, read on 25 September 2026 and again on 26 September 2026.
- NIST released the Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, on 26 January 2023, at https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf. MANAGE 3.1 was read in the published PDF on 25 September 2026 and again on 26 September 2026.
- DigitalXForce describes the TPRM module, External Risk View, X-ROC and its two AI engines in its glossary, read on 26 September 2026.
Do this once instead of every audit.
Every step above can be done by hand. DigitalXForce does them continuously, maps the result to 50+ frameworks once, and keeps the evidence current between audits. A 30 minute walkthrough on your own framework set shows what that removes from your calendar.
