AI is moving faster than your security controls. Are you still in control?

Secure AI adoption with e3 and Noma Security

Artificial Intelligence is rapidly becoming part of everyday business.

Employees are using generative AI. Developers are adopting coding assistants. Business teams are building their own copilots and agents. And increasingly, AI systems can access corporate data, call applications and perform actions autonomously.

The opportunity is significant.

But for CEOs and boards, AI also introduces a new management question:

Can we embrace AI at speed without losing control over our critical data, systems and business processes?

e3 has partnered with Noma Security to help organisations answer that question.

Noma provides a purpose-built enterprise AI Security platform for discovering, governing, testing and protecting AI applications and autonomous agents.

e3 combines this technology with specialist services around:

• AI Security strategy;
• AI risk assessment;
• AI governance;
• AI Security Posture Management;
• AI agent security;
• AI red teaming;
• runtime AI protection;
• Data Protection;
• DSPM and DLP integration;
• AI Security implementation;
• managed AI Security services.

Together, e3 and Noma help organisations enable AI without losing control.

AI is becoming part of your enterprise infrastructure

Until recently, AI was mainly something employees used through applications such as ChatGPT.

That is changing quickly.

AI is increasingly embedded in:

• Microsoft Copilot;
• Salesforce Agentforce;
• ServiceNow;
• software development;
• customer applications;
• data analytics;
• business workflows;
• cloud environments;
• autonomous AI agents.

AI is moving from:

answering questions

to:

taking actions.

An AI agent may be able to:

• search corporate information;
• read documents;
• access databases;
• call APIs;
• send messages;
• update applications;
• execute code;
• perform business transactions.

That means AI is no longer simply a productivity tool.

It is becoming part of your digital workforce and enterprise infrastructure.

And that requires a different security approach.

Five questions every CEO should be able to answer

For executive management, AI Security ultimately comes down to five questions:

1. Do we know which AI is operating within our organisation?
2. Do we know which critical information these AI systems can access?
3. Do we know what our AI agents are allowed to do?
4. Can we detect when an AI system behaves outside its intended boundaries?
5. Can we demonstrate to our board, customers and regulators that AI remains under control?

If the answer to any of these questions is unclear, there is likely an AI governance and security gap.

The e3 – Noma AI Security partnership

e3 has partnered with Noma Security to bring specialist AI Security capabilities to organisations adopting generative and agentic AI.

Noma is focused specifically on protecting the emerging AI and agentic enterprise.

Its platform is designed around four fundamental capabilities:

Discover → Govern → Test → Protect

Discover with AI Security Posture Management

Find AI applications, models, agents, MCP servers, tools and dependencies across the organisation.

Govern AI agents

Determine which agents are authorised, which information they may access and which actions they are allowed to perform.

Test AI continuously

Use AI-specific red teaming to identify vulnerabilities before they are exploited.

Protect AI at runtime

Detect and respond when AI applications or agents behave unexpectedly or unsafely.

Noma describes its platform around these areas of AI-SPM, Agent Access Control, AI Red Teaming and AI Detection & Response.

e3 adds the organisational, Data Protection and operational expertise required to turn those capabilities into a functioning enterprise AI Security programme.

Why e3 has partnered with Noma

AI introduces risks that traditional cybersecurity products were not originally designed to address.

Traditional security asks questions such as:

Is this user authorised?

Is this application vulnerable?

Is this device trusted?

AI introduces additional questions:

• Which AI agent is acting?
• On whose behalf?
• What information can it access?
• Which tools can it invoke?
• What happens when its behaviour changes?
• What happens if the model is manipulated?
• What happens if an MCP server or AI dependency is compromised?

e3 selected Noma because its technology is purpose-built around this new AI risk model.

This complements e3’s existing expertise in:

• Data Loss Prevention;
• DSPM;
• Insider Risk;
• Information Protection;
• data classification;
• AI Data Security;
• data sovereignty.

The combination enables e3 to approach AI Security from both sides:

protect the AI system

and

protect the data the AI system can reach.

You cannot govern AI you cannot see

One of the first AI Security challenges is visibility.

Many organisations do not have a complete inventory of the AI already operating within their environment.

AI may be introduced through:

• approved enterprise projects;
• individual business units;
• developers;
• SaaS platforms;
• browser applications;
• coding assistants;
• Copilots;
• autonomous agents;
• external suppliers.

This creates Shadow AI and increasingly Shadow Agents.

Noma AI Security Posture Management — AI-SPM — continuously discovers AI assets and their relationships.

This can include:

• AI agents;
• models;
• MCP servers;
• tools;
• skills;
• data connections;
• infrastructure.

Noma then maps these relationships to identify where AI systems may create significant risk.

From AI inventory to business risk

A list of AI applications alone is not particularly useful to a CEO.

The real question is:

What could this AI system actually reach or do?

Consider two AI agents.

Agent A can answer questions based on public product documentation.

Agent B can access:

• customer records;
• source code;
• financial systems;
• email;
• production infrastructure.

These agents clearly do not represent the same level of business risk.

Noma helps identify this blast radius by understanding how AI assets connect to data, tools, identities and applications.

e3 then helps translate those technical findings into:

• business impact;
• risk prioritisation;
• governance decisions;
• remediation actions.

Protect your crown jewels from AI

At e3, we believe AI Security should start with the organisation’s digital crown jewels.

These may include:

• intellectual property;
• source code;
• product designs;
• customer information;
• research data;
• pricing models;
• contracts;
• strategic plans;
• financial information.

The important question is not simply:

Which AI applications do we use?

It is:

Which AI applications and agents can reach our most critical information?

An AI system with access to sensitive data creates a new pathway through which that information can potentially be:

• exposed;
• copied;
• transformed;
• shared;
• misused.

This is why e3 integrates AI Security with Data Protection.

AI Security and Data Protection must converge

Many organisations currently treat AI Security as a separate project.

We believe that is a mistake.

AI interacts directly with data.

Therefore, AI Security must connect with:

• Data Loss Prevention;
• DSPM;
• classification;
• sensitivity labels;
• identity;
• access governance;
• Insider Risk;
• cloud security;
• data sovereignty.

For example:

DSPM determines where sensitive information resides.

Classification determines how important that information is.

DLP controls how information may be used or shared.

Identity determines who may access it.

Noma adds another critical layer:

Which AI agent can access it, and what can that agent do with it?

This gives organisations a much more complete picture of AI-related data risk.

Govern AI agents before they act

AI agents introduce one of the most important new enterprise security challenges.

Traditional identity governance asks:

What may this employee access?

Agentic AI adds:

What may an AI agent acting on behalf of that employee access and do?

Those permissions should not automatically be identical.

An employee may legitimately have access to sensitive customer information.

That does not necessarily mean every AI agent used by that employee should automatically receive the same access.

Noma’s Agent Access Control is designed to establish policies around AI-agent access and actions.

This enables organisations to define boundaries before an agent executes an action.

Give AI enough access — but not unlimited access

For example, an AI agent might be allowed to:

• retrieve selected customer information;
• summarise a document;
• generate a proposal;
• create a draft email.

But not to:

• download an entire customer database;
• export confidential information;
• modify production systems;
• execute financial transactions;
• send sensitive information externally.

This is the emerging principle of least privilege for AI agents.

e3 helps customers define and operationalise those policies.

MCP Security

Model Context Protocol — MCP — is rapidly becoming an important way to connect AI agents with applications, data and tools.

That makes MCP powerful.

It also makes it security-critical.

An MCP server may allow an AI agent to:

• retrieve corporate data;
• call an application;
• execute a task;
• access an API.

Noma’s platform is designed to discover and evaluate these connections and identify risks across AI agents, tools and their dependencies. Noma is also integrating its runtime security into broader AI gateway and agent ecosystems to inspect prompts, responses, tool calls and agent actions.

e3 helps organisations develop governance around the use of MCP and agent-to-tool connectivity.

Secure the AI supply chain

AI applications are rarely standalone.

They depend on combinations of:

• models;
• open-source components;
• libraries;
• MCP servers;
• tools;
• APIs;
• plugins;
• external services.

Together these create an AI supply chain.

A vulnerability or malicious component within that supply chain can affect multiple AI applications and agents.

AI Security must therefore extend beyond the AI model itself.

The organisation must also understand:

What does our AI depend on?

Which components do we trust?

What happens if one of those components changes?

Noma provides visibility across those AI dependencies, while e3 helps organisations incorporate these risks into their security and governance processes.

Test AI before attackers do

AI applications introduce vulnerabilities that traditional penetration testing may not identify.

Examples include:

• prompt injection;
• jailbreak techniques;
• sensitive-data leakage;
• agent manipulation;
• goal manipulation;
• unsafe tool calls;
• unintended agent behaviour.

Noma AI Red Teaming is designed to test AI applications and agents against AI-specific adversarial techniques.

Noma also aligns its AI security work with frameworks such as OWASP, MITRE ATLAS and industry AI security frameworks.

e3 helps organisations incorporate AI red teaming into their wider security testing process.

AI Security cannot stop at go-live

Testing an AI application before deployment is important.

But AI Security cannot finish when an application goes into production.

AI behaviour can change because:

• prompts change;
• data changes;
• models change;
• tools change;
• dependencies change;
• users interact with the AI differently;
• attackers manipulate its behaviour.

That means organisations need runtime visibility.

Noma AI Detection & Response

Noma AI Detection & Response — AI-DR — is designed to monitor AI behaviour during operation.

Rather than evaluating only individual events, Noma can correlate context across AI interactions.

This can help organisations detect:

• unexpected agent behaviour;
• malicious prompt activity;
• sensitive-data exposure;
• unusual tool use;
• policy violations.

Noma is increasingly integrating this runtime protection into AI gateways and agent platforms so controls can operate directly in the execution path.

Detect — and intervene

For autonomous AI, simply generating a security alert may not be sufficient.

An AI agent can act faster than a security analyst.

Controls may therefore need to:

• alert;
• block;
• mask sensitive information;
• prevent a tool call;
• restrict an action;
• require human approval.

The objective is:

give AI autonomy where it creates value, but maintain control where the risk becomes unacceptable.

The e3 AI Security Services

Technology is only one part of an effective AI Security programme.

e3 provides services across the complete AI Security lifecycle.

1. e3 AI Security Assessment

Understand your current exposure before designing controls.

The assessment examines areas including:

• AI applications already in use;
• Shadow AI;
• AI agents;
• AI development platforms;
• sensitive-data access;
• identities and permissions;
• AI supply chain;
• governance;
• existing security controls.

The result is a prioritised view of the organisation’s AI Security posture.

2. AI Security Strategy & Roadmap

e3 helps organisations translate technical findings into a practical roadmap.

We help answer:

• Which AI use cases create the greatest risk?
• Which controls are required?
• Where should Noma be deployed?
• How should AI Security integrate with existing security?
• Which capabilities should be implemented first?

The result is a phased approach aligned with business priorities rather than a technology-driven deployment.

3. Noma AI-SPM Implementation

e3 can design and implement Noma AI Security Posture Management.

This includes:

• platform architecture;
• integration;
• AI discovery;
• asset onboarding;
• risk prioritisation;
• dashboard design;
• operational processes.

The objective is to create a continuously maintained view of the AI environment.

4. Shadow AI & Agent Discovery

e3 helps organisations identify AI that may exist outside established governance.

This includes:

• unapproved AI applications;
• developer AI tools;
• AI agents;
• endpoint AI;
• SaaS agents;
• AI connections to sensitive systems.

This provides the foundation for meaningful AI governance.

5. AI Agent Security

As AI becomes more autonomous, Agent Security becomes increasingly important.

e3 helps organisations define:

• authorised agents;
• agent identities;
• access requirements;
• least privilege;
• allowed tools;
• prohibited actions;
• approval workflows.

Noma then provides technology to discover and enforce these controls.

6. MCP Security

e3 helps organisations evaluate the use of MCP within their AI architecture.

Services can include:

• MCP discovery;
• risk assessment;
• access review;
• tool and permission analysis;
• governance design;
• integration with agent-security controls.

7. AI Red Teaming

e3 can help organisations introduce AI-specific security testing using Noma.

Testing can focus on:

• prompt injection;
• jailbreaks;
• sensitive-data exposure;
• agent manipulation;
• unsafe tool usage;
• AI supply-chain vulnerabilities.

Findings can be translated into technical remediation and governance improvements.

8. AI Runtime Protection

e3 can implement and operationalise Noma AI Detection & Response.

This includes:

• runtime policies;
• detections;
• response actions;
• escalation;
• SOC integration;
• incident workflows.

This helps organisations move from periodic AI assessments towards continuous AI Security.

9. AI Data Security

AI Security must include protection of the information used by AI.

e3 combines Noma with our wider expertise in:

• DLP;
• DSPM;
• Microsoft Purview;
• Forcepoint;
• Proofpoint;
• Netskope;
• Zscaler;
• encryption;
• Insider Risk.

This enables us to design controls around both:

the AI

and

the information the AI uses.

10. AI Governance

Security technology needs clear governance.

e3 helps organisations define:

• acceptable AI use;
• ownership;
• roles and responsibilities;
• AI approval processes;
• security requirements;
• data-access policies;
• exception processes;
• incident procedures.

This helps turn an AI policy into operational controls.

11. AI Security & Sovereignty

AI introduces new dependencies on:

• cloud platforms;
• model providers;
• external AI services;
• APIs;
• data processors.

For organisations concerned with digital sovereignty, the question is therefore broader than where the data is stored.

It also includes:

Who controls the AI?

Who controls the underlying model?

Who can access the information?

Can the organisation change provider?

Can it continue operating if dependencies change?

e3 can incorporate AI into its broader Enterprise Sovereignty assessments.

12. Managed AI Security

AI environments will change continuously.

New models will appear.

New agents will be deployed.

New MCP servers will be connected.

New vulnerabilities will emerge.

That makes AI Security an ongoing process rather than a one-time implementation.

e3 can support organisations with managed AI Security services such as:

• continuous AI exposure review;
• Noma platform management;
• AI asset monitoring;
• security finding analysis;
• policy optimisation;
• risk prioritisation;
• incident support;
• AI red-team coordination;
• executive reporting.

From AI policy to demonstrable control

Many organisations already have an AI policy.

The challenge is proving that the organisation actually follows it.

We recommend moving through four stages:

Policy → Visibility → Control → Evidence

Policy

Define how AI may be used.

Visibility

Know what AI actually exists.

Control

Define what AI may access and do.

Evidence

Demonstrate that controls work.

e3 and Noma help turn AI governance from a document into an operational security capability.

The e3 AI Security Review

For organisations that are unsure where to start, e3 can perform an AI Security Review.

The review can assess:

AI Landscape

Which AI applications, platforms and agents are being used?

Crown Jewels

Which critical data may be exposed to AI?

Agent Access

Which actions and information can agents access?

Shadow AI

Where is AI being used outside established governance?

AI Supply Chain

Which models, MCP servers, tools and external dependencies exist?

Security Testing

Are AI applications tested against AI-specific threats?

Runtime Security

Can unsafe AI behaviour be detected and stopped?

Governance

Are roles, policies and responsibilities clearly defined?

What does the CEO receive?

The objective is not a technical report containing hundreds of AI findings.

Executive management receives clarity on:

• which AI is operating within the organisation;
• where the most important AI-related risks exist;
• which crown jewels AI can access;
• which AI agents have significant privileges;
• where governance gaps exist;
• which security controls should be prioritised;
• what the organisation should do next.

This results in a practical AI Security roadmap.

Why e3 and Noma?

Purpose-built AI Security technology

Noma focuses specifically on the security challenges created by AI applications and autonomous agents.

e3 Data Protection expertise

e3 has more than two decades of experience helping organisations protect their most sensitive information.

Technology + governance

We combine specialist technology with implementation, organisational processes and governance.

AI + Data

We do not treat AI Security and Data Protection as separate disciplines.

Vendor-independent architecture

e3 can integrate Noma with the wider security and Data Protection environment.

From assessment to operation

e3 can support the complete lifecycle:

Assess → Design → Implement → Govern → Test → Protect → Operate

The CEO perspective

The objective of AI Security should not be to slow down AI adoption.

Stopping AI is neither realistic nor desirable.

The objective is to make sure that increasing AI autonomy does not result in decreasing organisational control.

The board should therefore be able to answer:

• What AI do we have?
• What can it access?
• What can it do?
• How do we know when something goes wrong?
• Can we intervene?
• Can we prove that our controls work?

Enable AI — without losing control

AI will continue to move deeper into business processes.

Agents will become more autonomous.

And more sensitive information will become accessible to AI.

Organisations that can combine innovation with effective security and governance will be in a much stronger position to benefit from that change.

e3 and Noma help organisations adopt AI securely while retaining control over their critical data, systems and business processes.

• Discover your AI.
• Understand the risk.
• Govern your agents.
• Test continuously.
• Protect at runtime.
• Enable AI — without losing control.

CONTACT E3

Do you recognize your organization in our vision on data protection and DLP?
We would be happy to meet, learn more about your environment and explore how we can help protect your critical data.

Get in touch with e3 Benelux and let’s secure your business together.