Research area · GRC as an engineering discipline Agentic
Agentic GRC Engineering: Compliance as data, agents with guardrails
Frameworks are published as documents, mappings are created by hand, and the same requirements are implemented multiple times. Vami Research examines how GRC becomes an engineering discipline: with frameworks as machine-readable OSCAL data, a shared graph for management systems, and AI agents that take over routine work according to defined approval rules — with source references and an auditable trail.
- OSCAL
- Frameworks as Code
- GRC graph
- Integrated management system
- Autonomy levels L0–L3
- ISO/IEC 42001
- Trust Center
- Deep Dives
- 15
- Whitepapers
- 1
- Solutions
- 2
From PDF compliance to a queryable control landscape
NIS2, DORA, the CRA, and the AI Act do not call for isolated one-off measures but for lasting governance, risk, and management structures. Yet every framework has its own format and numbering scheme, and with every new version, the mapping work starts over. Organizations that build an ISMS, data protection, AI governance, and business continuity in parallel end up describing processes, collecting evidence, and answering audit questions multiple times.
The building blocks are in place. OSCAL turns controls into addressable data and, since version 1.2.0 of December 2025, has included a dedicated model for mappings between frameworks. Since September 29, 2025, the BSI has provided its State-of-the-Art Library on an OSCAL basis. The Harmonized Structure gives ISO/IEC 27001, 27701, 42001, and ISO 22301 the same ten clauses — the foundation for an integrated management system. One limitation remains: because of copyright, there is no freely distributable official OSCAL catalog for ISO 27001.
GRC becomes agentic when AI agents take over routine work on this foundation — traceably and under approval rules. In VamiGRC, VamiSec's agentic integrated management system, VamiAI answers questions from the IMS with source references; autonomy levels L0 to L3 determine what requires approval. Vamiset, usable standalone and currently without AI features, inventories assets in cloud, code, and HR with read-only permissions and checks them against controls. With “Powered by,” VamiSec's Trust Center identifies VamiGRC as its platform.
Open questions in practice
- 01
How does a framework PDF become an auditable, versionable control set?
Our approach
Through OSCAL: catalogs with stable IDs and verbatim text, profiles for selection and tailoring, mappings with justified relationship types. Version changes thus become diffs instead of fresh starts. In VamiGRC, the Framework Composer combines multiple standards into one control set, and your own frameworks are added via import — through to the Statement of Applicability.
- 02
How much autonomy may AI agents be given in governance, risk, and compliance processes?
Our approach
As much as the task can tolerate, and never unnoticed. In VamiGRC, every write operation runs at a defined autonomy level from L0 to L3: manual, AI-assisted, AI-executed with mandatory approval, or fully autonomous and fully audited. You decide per task where it runs, for example management reviews manually and continuous monitoring fully autonomously. The levels are inspired by OWASP APTS and adapted for GRC.
- 03
How do an AI assistant's answers remain verifiable in a GRC context?
Our approach
By having every answer point to a source and a human approve it before it is passed on. VamiAI answers questions from the IMS with source references and drafts notifications under NIS2, GDPR, DORA, the CRA, and the AI Act, which your team reviews and submits. For customer questionnaires, AI suggests answers from a versioned knowledge base; every answer keeps its audit trail with owner, source, and timestamp.
- 04
How does AI governance become part of the management system instead of a parallel project?
Our approach
Through an AI inventory with risk classification per use case and an AIMS per ISO/IEC 42001 that follows the same basic structure as ISO/IEC 27001: one asset register, one risk methodology, combined audits. The NIST AI RMF adds the risk perspective, and ISO/IEC 27090 the technical AI threats. In addition, the transparency obligations of the AI Act have applied since August 2, 2026.
Key takeaways
- 01
Frameworks as data turn multi-framework compliance into a data operation
OSCAL turns every control into an addressable object with a stable ID and verbatim text. Since version 1.2.0, mapping between frameworks has been part of the standard, with justified relationship types. Version changes can be assessed as a diff, and audits draw on verifiable references — provided a platform encapsulates the complexity of the standard.
Source: OSCAL — Frameworks as Code
- 02
Grundschutz++ arrives as an OSCAL catalog — auditing is meant to become automatable
According to its milestone plan, the BSI presented the methodology as BSI Standard 300-1 at it-sa on October 27, 2026. The Grundschutz++ user catalog bundles 1,000 requirements in 20 practices and is available in OSCAL format. According to the BSI, applications for certification to ISO 27001 based on Grundschutz++ are possible from January 1, 2027.
Source: Grundschutz++
- 03
AI security per ISO/IEC 27090 does not need a second management system
ISO/IEC 27090 is guidance without certifiable requirements. Its nine types of compromise feed into the ISMS risk assessment as a threat catalog, and its countermeasures into the Statement of Applicability. Internal audit, management review, and the certification audit cover them as well; certification continues to be to ISO/IEC 27001.
Source: ISO 27001 × ISO 27090
- 04
AI speeds up security questionnaires only with an answer library and review
AI-assisted tools recognize semantically similar questions and suggest draft answers from a curated library. This only becomes viable with governance: vetted answers with ownership and defined validity, an approval process, an audit trail per answer, and regular reconciliation with the actual state of controls.
Source: Security Questionnaires
How we work
We treat compliance as an engineering task: requirements as data, relationships as a graph, automation only with defined approvals — and every statement traceable to a source.
Frameworks as Code
Maintain frameworks as OSCAL catalogs and profiles, justify mappings between frameworks, and assess version changes as a diff.
Integrated management system
ISMS, PIMS, AIMS, CSMS, and BCMS with a shared risk scale, one document control process, and a combined audit program instead of parallel systems.
Agents with guardrails
Set autonomy levels per task, enforce approvals before consequential steps, and allow answers only with source references.
Continuous evidence
Inventory assets in cloud, code, and HR with read-only permissions and check them against controls ad hoc or on a schedule, for example with Vamiset and custom YAML rules, without AI features.
Trust Center and questionnaires
Provide evidence via self-service with tiered access, and answer customer questionnaires from a versioned answer library with review.
AI governance
AI inventory, risk classification per use case, and an AIMS per ISO/IEC 42001, complemented by the NIST AI RMF and the transparency obligations of the AI Act.
How to get started
- 1ModelConnect frameworks, contractual requirements, processes, assets, suppliers, AI use cases, and risks in one graph; implement requirements once and map them multiple times.
- 2AutomateHand routine work over to agents, each at a defined level: gap analyses and evidence drafts with mandatory approval, continuous monitoring fully autonomous and audited.
- 3DemonstrateMap evidence to controls, feed audits and management reviews from it, and demonstrate your status externally via a Trust Center.
Deep dives, whitepapers, and analyses
Deep Dives 15
From research area to implementation
Services
Related research areas
Set up GRC as an engineering discipline?
In an initial consultation, we assess your management systems and evidence sources and show what Agentic GRC Engineering with VamiGRC looks like in your environment.