Vortex AI Journey
AI governance and security in three stages: visibility, operation and protection. With the AI Journey Assessment, your company knows where the risk is, what regulation requires and where to start.
In short: Vortex AI Journey is Vortex Security’s offer for AI governance and security, from LLM applications to autonomous agents. It has three pillars: Visibility, with the AI Journey Assessment; Operation, with CAIO as a Service and the governance platform; and Protection, with AI protection solutions recommended without vendor bias and integrated into the Cyber Defense Center, the Converged NG-SOC (NG-SOC Convergente), which monitors the use and abuse of AI. The recommended starting point is the diagnostic.
The problem is not AI. It is AI without governance.
Three questions the board is already asking:
- 01 What AI is running today, approved or not?
- 02 Are we compliant with what regulation requires?
- 03 Who is accountable when a model exposes data or makes the wrong decision?
Three exposures that current controls do not cover
- 01 OWASP LLM03 · OWASP Agentic Top 10 The autonomous agent is a privileged user that nobody provisioned It uses a service credential, reads databases, calls APIs and executes transactions. It never went through onboarding, has no manager and does not show up in the access review.
- 02 OWASP LLM01 · MITRE ATLAS Prompt injection is remote execution with built-in social engineering A sentence in a document, email or web page that the agent reads becomes a command. The firewall does not see it, because the traffic is legitimate.
- 03 ISO/IEC 42001 · NIST AI RMF The certified ISMS sees neither of them ISO/IEC 27001 addresses confidentiality, integrity and availability. It does not address explainability, bias, agent drift or training traceability.
Source: IBM, Cost of a Data Breach Report 2026 (global press release of July 29, 2026, and IBM X-Force analysis).
Three pillars and what each one delivers
Each pillar answers a question and has its own component. Below is what your company gets in each one.
Visibility
Where are we?
AI Journey AssessmentA diagnostic with a proprietary methodology that pinpoints the gap and measures regulatory compliance.
What you get
- Maturity baseline by domain, across policy, control and enforcement
- Four KPIs for the board, with a risk heatmap
- Compliance map by framework, item by item
- Roadmap in three waves, with effort and cost per control
- AI architecture archetype and stack ranking
- Executive dashboard, PDF report and XLSX spreadsheet
- Board presentation and a prioritized plan that opens Pillar II
Operation
How do we govern?
CAIO as a Service and platformThe Pillar I plan becomes routine, with defined owners and cadence.
What you get
- On-demand executive AI leadership that owns the governance agenda
- Committee, roles, policy and acceptable use of AI
- Inventory of use cases, classified by risk, with an approval workflow
- Controls per use case
- Board reporting with metrics, and regulatory liaison
- Governance platform: a single register of models, agents, risks, approvals and evidence
- Evidence for audits and for ISO/IEC 42001
Protection
How do we protect?
Analysis and recommendationAnalysis of the environment and recommendation of the solutions that close each gap.
What you get
- Survey of the architecture, the data flows and the tools and licenses you already have
- Cross-check with the gaps prioritized in the assessment and with applicable regulation
- Definition of what needs to be controlled, what only needs adjustment and what is missing
- Evaluation of solutions without vendor bias
- Deployment and integration into the Cyber Defense Center
- Monitoring of AI use and abuse in the NG-SOC Convergente, together with IT and OT
Diagnose, govern, protect and measure again
The journey closes in a cycle. The new round of the assessment compares control by control and shows progress by framework.
The AI Journey pinpoints gaps and measures compliance.
CAIO and the platform carry out the plan.
Recommended solutions close the technical gaps.
A new round proves the progress.
You can start with any pillar. The recommended starting point is the diagnostic.
Eight scored domains and a Discovery, in four packages
A proprietary methodology, with domains designed to isolate the gap, with no generic questionnaire. Each package adds domains according to the company’s size and exposure. Any domain can be added later, without rework.
From smallest to largest: Express covers the foundation (D1 and D2) and the Discovery; Industrial covers everything, including AI in operational environments.
- Foundation Express
- D1 · AI governance: committee, roles, inventory and third parties
- D2 · AI risk: model risk and classification
- Express plus Standard
- D3 · Applications and LLM: OWASP LLM 2026, OWASP ML and RAG
- D6 · Privacy: legal basis, DPIA and international transfer
- Standard plus Enterprise
- D4 · Agentic AI: autonomous agents and MITRE ATLAS
- D5 · Secure SDLC and MLOps: pipeline and training data
- D7 · Continuity: BCP adapted to AI and dependence on external LLMs
- Enterprise plus Industrial
- D8 · Operational environments: IEC 62443 and IEC 61508
130 scored controls across eight domains, plus 32 Discovery questions. Two reading layers: executive, for the board; technical, with root cause and blocked controls.
AI maturity, exposure and readiness, measured with proprietary indices
Each control assessed in three independent dimensions
Policy
Is it written down?
An approved standard, with an owner, version and validity period. It is the easiest dimension to achieve and the one that protects least on its own.
Policy without control is a standard nobody carries out.
Control
Is it implemented?
A tool, process or automation that carries out what the policy sets out.
Control without enforcement is a tool switched on and never reviewed.
Enforcement
Does anyone check?
Auditable evidence that the control operates consistently over time.
Both situations can produce a high maturity score and fail the first audit.
Four patterns that a single score hides
In all four cases the average score can look good. It is the difference between the dimensions that shows where governance fails.
- Fails a technical audit Policy without control A flawless standard on paper. No tool covers the risk.
- Fails compliance Control without policy A tool in production with no formal approval and no declared owner.
- Fails accountability Tool without audit The control exists. Nobody measures whether it is working.
- Fails governance Audit without enforcement The evidence is collected, but the finding has no consequence.
How the score becomes a board decision
The score rolls up from the question to the organization, and each level feeds a view of the report. At the end, four indicators answer questions the board is already asking.
- Level 1
Question
(P + C + E) ÷ 3. Average of the three dimensions of each control.
- Level 2
Domain
0 to 100%. Average of the domain’s controls. Basis of the heatmap and the radar.
- Level 3
Organization
Domains in scope. Aggregation of the contracted domains. Basis of the executive dashboard.
- 0% to 25% Critical
- 25% to 50% High
- 50% to 75% Moderate
- 75% to 100% Low
- A
Governance maturity
Distance from ISO/IEC 42001: composite average of the controls mapped to the standard.
In the boardroom A prerequisite for certification and for compliance with the EU AI Act.
- B
Data exposure
Risk of leakage through external AI: cross-checks privacy compliance (LGPD, ANPD, GDPR) with actual AI use.
In the boardroom Sanctions from the ANPD, Brazil’s data protection authority, and reputational damage.
- C
Technical vulnerability
Adherence to the OWASP LLM Top 10: composite average of the mapped controls.
In the boardroom A cyber incident originating in a model.
- D
Business impact
Continuity risk: composite average of domain D7 and the ISO 22301 controls.
In the boardroom Operational loss and breach of business SLAs.
The Discovery never goes into a KPI or into maturity. Each KPI shows the percentage, risk classification, sample, universe, residual gap and distribution by domain.
From the diagnostic to a plan in three waves
Every control with a gap goes into the roadmap according to the nature of the control and its score, with effort, cost, owner and dependencies. The first wave is dominated by process: committee, policy, gate and contract.
Wave 1 · 0 to 60 days · Immediate protection
Critical items and blockers of other controls
Shadow AI, leakage via LLM and prompt injection
Removes the alerts that block the architecture recommendation
Wave 2 · 60 to 180 days · Structuring
Formal policies approved and communicated
Partial-coverage tools for the critical flows
DPIA per system and contracts with LLM providers
Wave 3 · 180 to 360 days · Evolution
Continuous monitoring, drift and bias
Audit with KRIs, integration with the ISMS and the AIMS
Preparation for certification, if applicable
Each gap becomes a work package
The plan reaches the investment committee with a deadline, owner and evidence of origin for each line.
- Short recommendation What to do, in one line.
- Path to the maximum score How the control reaches level 4 in all three dimensions.
- Type of action Process, tool or both.
- Estimated effort and cost Per control, aggregated by wave, domain and regulator.
- Action owner Responsible role, in vendor-agnostic language.
- Provenance Source framework, applicable regulator and the question that generated the finding.
What AI architecture should your company adopt?
The Discovery is the only part of the assessment that does not score maturity. It surveys need, constraints, data, scale and intended architecture, in a session with engineering and MLOps, and feeds a deterministic engine that indicates the architecture archetype, the alerts and the stack ranking.
A recommendation without vendor bias, traceable to the answer that generated it.
- 01 Discovery Need, constraints, data, scale and intended architecture.
- 02 Hard filters A tool that violates a mandatory requirement (regulated data, certification, residency) leaves the ranking, with the reason recorded.
- 03 Profile-based scoring The remaining tools are weighted according to the company’s profile: sector, sensitivity, latency, cost and sovereignty.
- 04 Archetype, alerts and stack ranking They go into the report, linked to the answer that generated them.
A critical alert blocks the recommendation until the prerequisite is resolved: the platform does not recommend a tool on top of an unaddressed risk. Example: an external LLM with sensitive data and no data processing agreement.
The architecture archetypes the Discovery indicates
The first three go from the lowest data sensitivity to the strictest residency. The other three serve specific uses: development, agents and machine learning.
The engine reports the status of each recommendation: stack ranked; decisive answers missing; catalog emptied by the filters; or blocking prerequisite, when a critical alert blocks the recommendation.
From kickoff to the board delivery
Reference cadence. The scope compresses to three sessions in Express and expands to ten or more in Industrial.
- Session 1 · Kickoff 2 h · Sponsor, CISO and DPO
Setup and scope
Segment, size, active domains, stakeholders and evidence channel.
All packages - Session 2 4 h
D1 and D2 · Governance and risk
Committee, roles, inventory, third parties and model risk.
All packages - Session 3 4 h
D3 and D6 · Applications and privacy
OWASP LLM 2026, OWASP ML, RAG, legal basis, DPIA and international transfer.
Standard and above - Session 4 4 h
D4 and D5 · Agentic and SDLC
MITRE ATLAS, autonomous agents, Secure SDLC, MLOps and training data.
Enterprise and above - Session 5 2 h
D7 · Continuity
BCP adapted to AI, dependence on external LLMs and contractual exit, with upload of the BIA and the BCP.
Enterprise and above - Session 6 · Optional 4 h
D8 · Operational environments
IEC 62443 zones and conduits and IEC 61508, with an OT specialist on site at the plant.
Industrial - Session 7 3 h · Engineering and MLOps
D9 · Discovery
Need, constraints, data, scale and intended architecture.
All packages - Consolidation 10 business days · offline
Processing and validation
Engines generate KPIs, waves, archetype and stack. Provenance review.
All packages - Final session 2 h · Board and C-level
Board delivery
Live dashboard, PDF and XLSX, wave plan with aggregated cost.
All packages - Follow-up Quarterly in Enterprise and Industrial; half-yearly in the others
Reassessment and delta
A new cycle with a before-and-after comparison by domain, KPI and control.
Recurring
What is included and what is not
Declared exclusions avoid surprises in the middle of the project. Out-of-scope items can be contracted separately, without rework, with full reuse of the baseline.
In scope
- Scored questions of the contracted domains, in all three dimensions
- The Discovery questions (D9)
- Collection and attachment of the evidence provided by the company
- Three-dimensional score and hierarchical aggregation
- Four KPIs and risk heatmap
- Regulatory crosswalk applicable to the segment
- Roadmap in three waves with effort and cost per control
- Architecture archetype and filtered stack ranking
- Executive report in PDF and analytical spreadsheet in XLSX
- Board presentation
- Hosting of the assessment and the evidence for 12 months
Not in the scope of the assessment
- Carrying out the remediation of controls with gaps
- Drafting corporate policies (the library templates are included in the Enterprise and Industrial packages)
- Deployment, configuration or licensing of tools: that is Pillar III, Protection
- Penetration testing or red teaming of models (add-on module)
- Statement of Applicability and preparation for Stage 1 and 2 (add-on module)
- Coordination with the certification body
- Legal opinion on regulatory classification
- Formal DPIA per AI system
- Automated technical inventory of models in production
- Source code audit of models or pipelines
- Domains and extensions not contracted
CAIO as a Service
On-demand executive AI leadership that owns the governance agenda. For companies that need to structure this agenda without necessarily creating a new internal executive position.
- Leads the AI committee and policy
- Prioritizes use cases by risk
- Reports to the board with metrics
- Handles regulatory liaison
How governance is put in place
The Pillar I plan becomes routine, with defined owners and cadence.
Four levels, at the cadence the moment calls for
The level does not dilute scope: it defines cadence, seat at the table and squad composition. All of them run governance inside the platform; what changes is the depth of the operation.
Advisory
Emerging governance. AI is relevant but still has no formal owner.
2 days a month- Squad
- Remote senior CAIO
- Seat
- Guest on the AI committee
- Curation of use cases and a shadow AI gate
- Policy review and regulatory updates
- Half-yearly reassessment in the AI Journey
Embedded
Waves 1 and 2 under way. An auditor or regulator has already asked.
1 day a week- Squad
- CAIO and GRC analyst
- Seat
- Member of the AI committee
- Execution of Wave 1, policy authoring and DPIA per system
- Review of contracts with LLM providers and guardrail tuning
- Quarterly reassessment with a live dashboard
Executive
AI in a critical flow or in automated decision-making. The board is demanding answers.
2 to 3 days a week- Squad
- CAIO, GRC and MLOps
- Seat
- Executive, reporting to the board
- Preparation for ISO 42001 Stage 1 and Stage 2, and red teaming
- Governance of autonomous agents
- Quarterly reassessment with the delta presented to the board
Industrial
AI at the plant: a model in the control loop or predictive maintenance.
2 to 3 days a week- Squad
- CAIO, GRC, MLOps and OT
- Seat
- Executive, with liaison with engineering
- The full Executive scope
- Dedicated OT specialist (IEC 62443)
- Zones and conduits, and IEC 61508 functional safety
A single register of what exists, who approved it, what the risk is and what the evidence is
CAIO and the platform carry out the plan that came out of the diagnostic. The platform keeps what exists, who approved it, what the risk is and what the evidence is for each use case.
Item-by-item adherence
Percentage met per framework, with the requirement, evidence and gap of each control.
Traceability
Each recommendation shows the question and the answer that generated it, with the regulatory requirement.
Delta on reassessment
The new round compares control by control and shows progress by framework.
Five modules that operate what the assessment measured
The platform does not replace the consultant: it gives them leverage. Manual tasks start running in the system, with the CAIO leading instead of operating spreadsheets.
Prioritization and inventory
Module 1 · D1 GovernanceA scattered backlog becomes a portfolio: each initiative with stage, investment, risk and owner, in an impact × complexity matrix.
AI policy
Module 2 · D1 GovernanceCreation with regulatory context, traceable approval, versioning and periodic review.
Agent governance
Module 3 · D3 and D4Monitoring, FinOps per agent, structured logs, guardrails, human validation and failover; model benchmarking by quality, latency and cost.
Risk management
Module 4 · D2 and D6Centralized inventory, risk classification anchored in PL 2338/2023 (Brazil’s AI bill), mitigation, review and sign-off.
Training and culture
Module 5 · Cross-cutting enforcementLearning paths, a curated contributions hub and gamification. Keeps the policy alive.
- Reading mode Board and executive The board and C-level see KPIs, heatmap, radar and roadmap, read-only.
- Reading mode Consultant Vortex and the company’s technical team see everything.
- Reading mode Guest A critical supplier or remote unit fills in only the domains released to it, through a temporary link, with access logged.
How we reach the recommendation
The analysis starts from what the company already has, to make the most of the investments already made and apply new controls only where there is a real gap.
| Capability assessed | What it solves |
|---|---|
| Usage discovery and control | Visibility of shadow AI and access policies |
| Data protection | DLP for AI prompts, uploads and responses |
| Application and model security | Defense against prompt injection, abuse and exposure |
| Identity and access | User and agent permissions to data and systems |
| Infrastructure posture | Configuration of cloud, workloads and AI pipelines |
| Governance, risk and compliance | Inventory, risk register and evidence |
| Monitoring and response | AI telemetry correlated in the CDC |
The recommendation takes into account what the company already has and the Vortex partner portfolio.
IT + OT + AI in the same security context
Signals of AI use and abuse enter the NG-SOC Convergente together with IT and OT. Detection gains context, and response follows the CDC playbooks.
- AI telemetry correlated with IT and OT
- Detection of leakage and abusive use of AI
- Response through the CDC playbooks
Data leakage via AI: AI, Customer data sent to a generative AI app; IT, Unusual export of the customer database. Response through the CDC playbook.
Abusive use of AI: AI, AI agent querying data outside its scope; OT, Bulk reading of production data from the plant. Response through the CDC playbook.
Security and governance aligned with your regulatory environment
The mapping is done control by control, with requirement, evidence and gap, which creates traceability for audits.
- AI governance
- ISO/IEC 42001: certifiable AI management system
- NIST AI RMF: govern, map, measure and manage risk
- EU AI Act: obligations by risk level
- PL 2338/2023: the bill for Brazil’s legal framework for AI
- Security
- ISO/IEC 27001 and NIST CSF 2.0: information security baseline
- OWASP Top 10 for LLM Applications: risks of applications built on language models
- OWASP Top 10 for Agentic Applications: risks of autonomous agents
- MITRE ATLAS: attack tactics against AI systems
- Privacy
- LGPD (Brazil’s General Data Protection Law): personal data in training, prompts and decisions
- GDPR: operations involving European data subjects
- ISO/IEC 27701: privacy management
- Industrial environments
- IEC 62443: security of automation systems
- IEC 61508: functional safety
- NERC CIP: critical energy infrastructure
Sector regulations come in according to the company’s segment (Brazilian regulators such as BACEN, SUSEP, ANS and ANEEL, among others).
Start with the diagnostic or contract the full journey
Three ways to engage, depending on where your company stands.
- Model 1 · Starting point
Diagnostic
AI Journey Assessment with dashboard and roadmap.
- Format
- One-off project
I · Visibility - Model 2
Diagnostic and governance
Assessment and CAIO as a Service.
- Format
- Recurring
I · Visibility II · Operation - Model 3
Full journey
Assessment, CAIO, platform, analysis and managed protection.
- Format
- Recurring with managed services
I · Visibility II · Operation III · Protection
Frequently asked questions
What does Vortex deliver in the AI Journey Assessment?
A maturity baseline by domain, across policy, control and enforcement; four KPIs for the board, with a risk heatmap; a compliance map by framework; a roadmap in three waves with effort and cost per control; an architecture archetype and stack ranking; an executive dashboard, a PDF report and an XLSX spreadsheet; and the board presentation.
Where should I start? Do I need to contract the whole journey?
No. You can start with any pillar, and the recommended starting point is the diagnostic. There are three ways to engage: Diagnostic; Diagnostic and governance; and Full journey.
Can I contract only some domains?
Yes. The domains are grouped into four packages, Express, Standard, Enterprise and Industrial, according to the company’s size and exposure. Any domain can be added later, without rework.
Does the assessment include penetration testing or red teaming of models?
No. Penetration testing and red teaming of models are add-on modules, contracted separately. The assessment evaluates each control in policy, control and enforcement, based on the working sessions and the evidence provided by the company. Detection and response controls undergo technical validation by the Cyber Defense team, and declared controls are probed by the Offensive team; active testing stays in the add-on module.
What is the Discovery?
It is the block of questions about need, constraints, data, scale and intended architecture. It does not score maturity and never goes into the KPIs: it feeds the architecture recommendation and the stack ranking.
Does Vortex deploy the recommended tools?
Not in the assessment: deployment, configuration and licensing are out of scope. Deployment is part of Pillar III, Protection, which evaluates solutions without vendor bias, deploys them and integrates them into the Cyber Defense Center.
Does AI protection require buying new tools?
Not necessarily. The analysis starts from what the company already has: architecture, data flows, tools and licenses. It cross-checks that survey with the gaps prioritized in the AI Journey and defines what needs to be controlled, what only needs adjustment and what is missing, so that new controls are applied only where there is a real gap.
What is CAIO as a Service?
It is on-demand executive AI leadership that owns the governance agenda. It leads the AI committee and policy, prioritizes use cases by risk, reports to the board with metrics and handles regulatory liaison, without the company having to create a new internal executive position.
Do I need to be using AI already to do the assessment?
The assessment assumes at least one AI system in operation, pilot or evaluation. Without that, it is still valid, but the applications, agentic AI and development domains tend to score by absence.
Which frameworks and regulations are considered?
ISO/IEC 42001, NIST AI RMF, EU AI Act, PL 2338/2023, ISO/IEC 27001, NIST CSF 2.0, OWASP for LLMs and for agents, MITRE ATLAS, LGPD, GDPR, ISO/IEC 27701 and, in industrial environments, IEC 62443, IEC 61508 and NERC CIP, as well as sector regulations such as BACEN, SUSEP, ANS and ANEEL.
How does AI fit into the Cyber Defense Center monitoring?
Signals of AI use and abuse enter the Vortex NG-SOC Convergente together with those from IT and OT. Detection gains context, and response follows the CDC playbooks.
Does Vortex do generative AI and agent security?
Yes. Vortex AI Journey covers the security and governance of LLM applications and autonomous agents. The AI Journey Assessment evaluates the applications and LLM domain (OWASP LLM 2026, OWASP ML and RAG), from the Standard package up, and the agentic AI domain (autonomous agents and MITRE ATLAS), from Enterprise up. The governance platform has an agent governance module, and the Protection pillar evaluates solutions for defense against prompt injection, abuse and exposure and for controlling user and agent permissions to data and systems, with AI telemetry correlated in the CDC. The concepts, the risks and the guardrails are in the AI security guide.
What does Vortex deliver in guardrails?
On three fronts. In the governance platform, the agent governance module brings together monitoring, structured logs, guardrails, human validation and failover. In CAIO as a Service, the Embedded level includes the review of contracts with LLM providers and guardrail tuning. And in the Protection pillar, Vortex evaluates, without vendor bias, the solutions that apply these controls (DLP for AI prompts, uploads and responses, defense against prompt injection, abuse and exposure, and user and agent permissions to data and systems), deploys them and integrates them into the Cyber Defense Center.
How does Vortex help control shadow AI and data leakage through AI?
The assessment measures data exposure, one of the four KPIs for the board, which cross-checks privacy compliance (LGPD, ANPD, GDPR) with actual AI use. In the roadmap, shadow AI, leakage via LLM and prompt injection go into the first wave, from 0 to 60 days. In the Protection pillar, Vortex evaluates usage discovery and control (visibility of shadow AI and access policies) and data protection (DLP for AI prompts, uploads and responses). And the detection of leakage and abusive use of AI goes into the NG-SOC Convergente, correlated with IT and OT. How the SOC itself uses AI is in the guide AI in the SOC.
Is your company ready to evolve with the use of AI?
Find out your organization’s AI maturity level, identify the main gaps and understand which actions should be prioritized to evolve with more security and control. The recommended starting point is the AI Journey Assessment.
Updated on