Skip to content
Nueva Cyber · Secure AI

Put AI to work without opening a new way in.

Copilots, chat tools and AI agents now read your email, search your files and act inside your systems. Nueva Cyber secures how your organisation uses, builds and runs AI, governed against NIST and ISO/IEC 27001 and tested the way attackers will test it.

Incident happening now? +61 2 7252 5678

AI exposure checkExample
  • SHADOWUnapproved AI tools in use · 6Found
  • COPILOTOversharing · finance site open to allFix
  • AGENTService identity with write accessReview
  • POLICYAI use policy · approvedDone
Illustrative example · Govern, protect, test, operate
  • ISO/IEC 27001 certified
  • SMB1001 Gold
  • Nueva 24x7 SOC
  • In house digital forensics and incident response
  • CREST Registered Testers in our team
  • ARN Innovation Awards: SMB Partner of the Year 2024 and 2025

ISO/IEC 27001 certified · Nueva 24x7 SOC · CREST Registered Testers · AI governance aligned to NIST and ISO/IEC 42001

See the Nueva Secure AI framework
The Challenge

AI is already inside your business. The question is who is governing it.

Most organisations did not decide to adopt AI. It arrived in the tools they already pay for, in browser extensions staff installed themselves and in projects a single team started. The result is AI with broad access to data and systems, and very little oversight of what it can see or do.

What we see in almost every environment

AI you cannot see.

Staff paste customer data, contracts and code into public AI tools. Nobody has a list of which tools are in use or what they retain.

AI that sees too much.

Enterprise copilots respect existing permissions. Years of oversharing in SharePoint, Teams and OneDrive mean an assistant can surface salary files or board papers to anyone who asks the right question.

AI that can be talked into things.

Prompt injection lets a malicious email, web page or document instruct an AI system to leak data or take an action. It is the top risk in the OWASP Top 10 for LLM Applications.

AI that acts on its own.

Agents hold credentials, call APIs and change records. An agent with more access than it needs is a privileged account nobody is watching.

AI you inherited from a supplier.

Models, plugins and SaaS features arrive with their own data flows, hosting locations and training terms.

AI your board will be asked about.

Directors are expected to understand material technology risk. Australia's Privacy Act transparency obligations for automated decisions commence on 10 December 2026, and customers increasingly ask about AI in due diligence and tenders.

None of these risks is solved by a policy document alone, and none is solved by a single product.

How Nueva Helps

Govern it, protect it, test it, watch it.

We treat AI as what it is: a new class of user, application and supplier, all at once. So we secure it the same way we secure everything else, starting with the business risk and working down to the control.

Govern.

Decide what AI the organisation will use, who owns each risk and how decisions reach the board. We build this on the NIST AI Risk Management Framework and ISO/IEC 42001 controls, and run it inside an ISO/IEC 27001 management system, so AI governance is part of the security program rather than a separate binder.

Protect.

Fix the data access, identity and architecture problems that make AI dangerous. Most of the work sits in permissions, classification and agent access, not in the AI tool itself.

Test.

Attack your AI applications, copilots and agents the way a real adversary would, using OWASP and MITRE ATLAS techniques, and prove the guardrails hold.

Operate.

Bring AI activity into the same 24x7 monitoring as your endpoints, identities and cloud, so misuse and AI enabled attacks are detected and contained.

The team that finds a gap can also close it. Nueva Technology remediates permissions, labelling and configuration, and Nueva Cyber keeps the controls under watch once they are in place.

Secure AI services

AI Security Readiness Assessment

The question it answers

Where are we exposed today, and what do we fix first?

What you receive

Current state against the Nueva Secure AI Framework, AI risk register, prioritised roadmap, executive briefing

Shadow AI Discovery and Control

The question it answers

Which AI tools are staff actually using, and with what data?

What you receive

AI usage inventory, risk rating per tool, approved tool list, enforced controls through your SSE, CASB or browser controls

Copilot and Enterprise AI Readiness

The question it answers

Is our data safe to put an AI assistant on top of?

What you receive

Oversharing analysis across SharePoint, Teams and OneDrive, sensitivity labelling and DLP design, remediation by Nueva Technology, rollout guardrails

AI Governance and Policy

The question it answers

Who owns AI risk, and what are the rules?

What you receive

AI policy and acceptable use standard, AI system register, roles and approval workflow, board reporting pack, delivered with our GRC team

AI Threat Modelling and Secure Architecture

The question it answers

Is the AI application we are building safe by design?

What you receive

Threat model for your GenAI, RAG or agent architecture, attack paths mapped to MITRE ATLAS, control design, architecture review

AI and LLM Security Testing

The question it answers

Can our AI be manipulated, jailbroken or turned against us?

What you receive

Testing against the OWASP Top 10 for LLM Applications and the OWASP Top 10 for Agentic Applications, prompt injection and data leakage findings, retest after fixes

AI Agent and Machine Identity Security

The question it answers

What can our agents reach, and who approves what they do?

What you receive

Agent and service account inventory, least privilege redesign, token and secret handling, human approval points for high impact actions

Managed AI Security Monitoring

The question it answers

Would we know if AI was being misused or used against us?

What you receive

AI activity telemetry into our 24x7 SOC, detections for misuse, data exfiltration and AI enabled attacks, AI incident playbooks

Private AI environments. For organisations that cannot send sensitive data to public AI services, Nueva Cyber and Nueva Technology design and secure private AI running on your own infrastructure, with the same framework applied from day one.

One framework for our own AI and yours.

We secure AI with the same framework we use inside Nueva. It joins recognised standards to practical controls, so every policy decision traces to a control, and every control can be tested and monitored. Governance comes from the NIST AI Risk Management Framework and the AI management controls in ISO/IEC 42001. Management discipline comes from ISO/IEC 27001, the standard Nueva is certified to. Technical controls come from the sources security practitioners already trust.

Four layers turn AI principles into controls you can test

Govern · NIST AI RMF and ISO/IEC 42001 controls

Govern, Map, Measure, Manage · ISO/IEC 42001 Annex A controls · GenAI Profile (NIST AI 600-1) · Australian AI6 practices

Read it top down: NIST and ISO/IEC 42001 set what good AI governance looks like, ISO/IEC 27001 makes it a managed and audited system, the eight domains are where controls live, and the technical references define how each control is built and tested.

Control domains and mapping

DomainKey controlsNIST AI RMFISO/IEC 42001 Annex AISO/IEC 27001:2022 Annex ATechnical references
1. Governance and accountabilityAI policy and acceptable use, named risk owners, approval workflow for new AI use, AI impact assessment, board reporting, staff awarenessGovernA.2, A.3, A.55.1, 5.2, 5.31, 6.3Australian AI6 practices
2. AI inventory and shadow AIRegister of AI systems, models, agents and plugins; discovery of unsanctioned tools; risk rating per systemMapA.4, A.95.9, 5.10NIST AI 600-1
3. Data protection for AIClassification and labelling, oversharing remediation before copilot rollout, DLP on prompts and uploads, retention, personal information handlingMap, ManageA.75.12, 5.13, 5.34, 8.11, 8.12OWASP LLM02, LLM08; ASD AI data security
4. Identity and agent accessLeast privilege for agents and service accounts, scoped short lived tokens, phishing resistant MFA for AI admins, human approval for high impact actionsManageA.95.15, 5.18, 8.2, 8.5OWASP LLM06; OWASP Agentic Top 10
5. Secure AI build and architectureThreat modelling, input and output validation, system prompt hygiene, guardrails, RAG data isolation, secrets out of promptsMap, ManageA.68.25, 8.26, 8.27, 8.28OWASP LLM01, LLM05, LLM07; ASD Guidelines for secure AI system development
6. AI supply chain and third partiesVendor due diligence, model and dataset provenance, hosting location, contract terms on training use of your dataGovern, MapA.105.19, 5.20, 5.21, 5.22, 5.23OWASP LLM03, LLM04; ASD AI and ML supply chain guidance
7. Assurance testingAI red teaming, LLM and agent penetration testing, guardrail evaluation before release and after changeMeasureA.68.8, 8.29MITRE ATLAS; OWASP LLM and Agentic Top 10
8. Detect and respondLogging of prompts, outputs and agent actions; anomaly and abuse detection; usage and cost limits; AI incident playbooks; kill switch for agentsManageA.6, A.85.24 to 5.28, 8.15, 8.16OWASP LLM10; NIST Cyber AI Profile (draft)

Baseline controls beyond the standards

Standards tell you what to manage. These twelve controls are the practical baseline we apply and recommend, because they close the gaps attackers use first.

  1. Approve AI tools before use, and block or coach on everything else at the browser and network edge.
  2. Keep a live register of every AI system, agent and plugin, with a named owner.
  3. Fix oversharing and apply sensitivity labels before switching on any enterprise copilot.
  4. Stop sensitive data leaving in prompts and uploads with DLP on sanctioned and unsanctioned AI.
  5. Give every agent its own identity with least privilege, short lived credentials and no shared accounts.
  6. Require a human to approve any AI action that moves money, changes access, deletes data or contacts customers.
  7. Treat all AI input as untrusted, including email, documents and web content an AI reads on a user's behalf.
  8. Validate and constrain AI output before it reaches another system, a database or a customer.
  9. Log prompts, responses and agent actions to the SOC, with retention that supports investigation.
  10. Set usage, rate and spend limits on every model endpoint and API key.
  11. Test AI applications and agents before go live and after every material change.
  12. Maintain an AI incident playbook and a tested way to disable any agent immediately.

Customers pursuing certification. The framework already draws on ISO/IEC 42001 controls, so a customer that intends to certify an AI management system carries this work straight into the certification effort.

How It Works

From first conversation to AI you can defend.

STEP 01 · DISCOVER.

We find the AI already in use: sanctioned tools, shadow AI, copilots, agents and AI features inside your SaaS platforms. You see what exists, who owns it and what data it touches.

STEP 01 OF 06

Where to start Most organisations begin with the AI Security Readiness Assessment. Organisations about to roll out a copilot start with Copilot and Enterprise AI Readiness. Teams building their own AI application start with AI Threat Modelling.

Outcomes

What changes when AI is secured properly

Adoption goes faster, not slower.

Clear rules and approved tools replace blanket bans and quiet workarounds.

Your data stays where it belongs.

Copilots and AI tools only reach what each person is entitled to see.

Agents do only what they should.

Every agent has an owner, a defined scope and a human check on high impact actions.

Weaknesses are found in testing, not in an incident.

Prompt injection and data leakage paths are fixed before go live.

Misuse is detected and contained.

AI activity is monitored alongside the rest of your environment, every hour of every day.

You can answer the hard questions.

Boards, auditors, insurers and customers get evidence mapped to NIST and ISO/IEC 27001, not assurances.

Why Nueva

Why organisations secure AI with Nueva

We run the framework we recommend.

Nueva is certified to ISO/IEC 27001, and we govern our own AI use with the same framework we apply for customers, drawing on NIST and ISO/IEC 42001 controls.

We test as well as advise.

Our Offensive Security team holds individual CREST Registered Tester, OSCP, OSWE and OSEP certifications and tests AI applications and agents alongside networks, cloud and identity.

We fix what we find.

Most AI risk sits in permissions, data and configuration. Nueva Technology remediates it, so findings do not stall in a report.

We watch it after go live.

AI telemetry flows into our Nueva 24x7 security operations centre, staffed by Nueva analysts.

We start from risk, not a product.

We work across Microsoft Security, SentinelOne, CrowdStrike, Netskope, Wiz and other leading platforms, and recommend the capability you need before the technology.

We know the Australian context.

Our advice reflects ASD guidance, the Essential Eight, the Privacy Act and Australia's Guidance for AI Adoption.

FAQ

Do we need to stop staff using AI while we sort this out?

No. Blanket bans push usage underground. We usually approve a safe set of tools quickly and tighten controls as the assessment progresses.

Is Microsoft 365 Copilot secure?

Copilot works within your existing permissions. If those permissions are too broad, Copilot makes that exposure easy to find. Fixing oversharing and labelling first is the most important step.

How is AI security testing different from a penetration test?

It targets how an AI system behaves, not just the infrastructure around it: whether it can be instructed to leak data, bypass its rules or misuse the tools and access it has been given.

Which framework should we use?

For most Australian organisations, the NIST AI Risk Management Framework for governance, run inside an ISO/IEC 27001 aligned management system, gives the strongest foundation. Adding ISO/IEC 42001 controls, as we do internally, gives you a clear path to certification if you want it.

We are building our own AI application. When should we involve you?

At design. A threat model early in the build is far cheaper than redesigning after a failed test.

Find out where your AI is exposed.

Start with an AI Security Readiness Assessment and leave with a clear view of your AI use, your gaps and the order to fix them.