2028: The AI-Native Business and the New Definition of Cybersecurity

This article is a forward-looking scenario in the DotlyGuard AI and cybersecurity series. It explores what an AI-native business could look like if the trends discussed in the previous chapters continue. It is not a prediction that every organization will adopt this model by 2028.

This article is Part 6 of From Generative AI to AI Agents: Why the Next Five Years Will Change Cybersecurity.

In 2023, artificial intelligence became accessible to millions of people through generative AI.

In 2024, AI increasingly entered business applications and workflows.

In 2025, AI agents began moving from generating responses toward performing tasks.

In 2026, the conversation expanded from individual agents to the agentic enterprise.

In 2027, the next possible step was the multi-agent organization, where multiple specialized agents could coordinate work.

Now consider the next question:

What happens when a business is designed around AI from the beginning rather than adding AI to an existing business?

That is the idea of the AI-native business.

An AI-native business would not simply use AI as another software feature.

AI could become part of the operating architecture itself.

People would still exist.

Customers would still exist.

Employees would still use computers.

Business applications would still exist.

But many operational processes could be coordinated by AI systems working across applications, data, APIs, and specialized agents.

That possibility changes cybersecurity once again.

The question is no longer only:

How do we secure AI?

It becomes:

How do we secure a business whose operating environment increasingly includes AI?

The progression from 2023 to 2028 can now be viewed as one continuous transformation.

| Year | AI role | Primary security question | | --- | --- | --- | | 2023 | AI generates | What information are we giving AI? | | 2024 | AI integrates | What systems can AI access? | | 2025 | AI acts | What can AI do? | | 2026 | AI operates across workflows | How do we secure an organization where AI can act? | | 2027 | AI agents coordinate | How do we secure relationships between agents? | | 2028 | AI becomes part of business architecture | How do we secure an AI-native organization? |

The important part is the progression.

AI is moving from a feature toward an operational component.

The security boundary moves with it.

An AI-native business is different from a traditional company that simply adopts AI tools.

A traditional business might look like:

Employees → Applications → Business Processes

AI is added:

Employees → AI Tools → Applications → Business Processes

An AI-native architecture could eventually look more like:

People + AI Agents + Applications + APIs + Data + Automated Workflows

The distinction is architectural.

AI is no longer an isolated tool.

It becomes part of how work is coordinated.

For example, imagine a company receiving a new customer inquiry.

A traditional process might be:

1. Employee receives inquiry.

2. Employee researches the customer.

3. Employee checks availability.

4. Employee prepares a proposal.

5. Employee enters the information into the CRM.

6. Employee schedules follow-up.

7. Employee sends the response.

An AI-enabled process might involve agents performing several of these tasks.

An AI-native process could be designed around those agents from the beginning.

The system might:

1. Receive the inquiry.

2. Identify the customer.

3. Research relevant information.

4. Determine the appropriate workflow.

5. Consult specialized agents.

6. Retrieve internal data.

7. Prepare the response.

8. Update business systems.

9. Schedule follow-up.

10. Escalate exceptions to a human.

The human may still make the important decisions.

But the operating model is different.

In traditional software architecture, the application layer coordinates business operations.

In an AI-native architecture, AI systems may increasingly participate in that coordination.

That means AI could become part of the business control plane.

Consider:

Customer → AI Interface → Coordinator Agent → Specialist Agents → Business Applications → Data → Actions

The coordinator determines what needs to happen.

Specialist agents perform individual tasks.

Applications execute operations.

Data systems provide information.

People intervene where required.

This creates an environment where the software is not simply executing fixed instructions.

It may be interpreting goals and deciding which permitted operations to perform.

That distinction has major security implications.

Traditional automation generally follows predefined paths.

For example:

When order received → create record → send notification

AI-native operations could involve greater adaptability:

Understand order → determine required steps → select appropriate tools → retrieve information → evaluate results → perform authorized actions

The second model is more flexible.

But flexibility creates another question:

Who controls the decision-making boundaries?

The organization needs policies that determine:

• What agents may do

• What data they may access

• Which tools they may use

• Which actions require approval

• Which actions are prohibited

• How long authority remains valid

• How activity is monitored

• What happens when an agent encounters an unexpected situation

The security architecture becomes part of the operating architecture.

Historically, organizations often treated security as a layer added around applications.

The application performed the business operation.

Security protected the application.

In an AI-native organization, security may need to become part of the workflow itself.

Consider a simple operation:

Agent → CRM → Update Customer

The security system needs to determine:

Who is the agent?

Why is it making the request?

What customer is involved?

Does the agent have permission?

Is the requested change within policy?

Was the request initiated by an authorized user?

Should the operation require approval?

How will the operation be recorded?

Security therefore becomes part of the action path.

The traditional enterprise identity model has users, groups, applications, and service accounts.

An AI-native business may introduce another class of identity:

AI agents.

But an agent identity alone is not enough.

Organizations may need to distinguish between:

• Human identity

• Device identity

• Application identity

• Service identity

• Agent identity

• Delegated identity

The same person may interact with multiple agents.

One agent may operate on behalf of another.

An agent may perform work on behalf of a human.

An application may invoke an agent.

The identity chain becomes more complicated.

For example:

Employee → Agent A → Agent B → Application → Database

The organization needs to preserve the relationship between these identities.

Otherwise, accountability becomes difficult.

NIST's work on AI agent identity and authorization reflects this direction, including the need for identification, authentication, authorization, auditing, and accountability for agent interactions (NIST comments on software and agentic AI identity).

It would be easy to assume that AI-native businesses eliminate traditional endpoints.

They do not.

People still use:

• Laptops

• Desktops

• Browsers

• Mobile devices

• Developer environments

• Administrative workstations

Even highly automated businesses still have humans.

Those humans interact with the AI operating environment through endpoints.

An administrator might approve a high-risk action.

A developer might configure an agent.

A security analyst might investigate an alert.

An employee might review an AI-generated customer response.

A manager might approve a financial operation.

Every one of these activities can involve an endpoint.

This means endpoint security remains foundational even in an increasingly autonomous business.

Consider a developer working on an AI-enabled application.

The developer's endpoint may have access to:

• Source code

• Cloud credentials

• Development tools

• AI coding agents

• Internal repositories

• Deployment systems

A compromise of that endpoint could therefore affect more than the developer's workstation.

It could potentially become a gateway into the organization's AI operating environment.

This is why endpoint visibility remains important.

Security teams need to understand:

What devices exist?

Who uses them?

What applications are installed?

What security activity is occurring?

Which devices interact with sensitive systems?

Which devices represent important operational paths?

AI does not make these questions obsolete.

It can make their answers more important.

Traditional security often divides environments into assets.

Servers.

Endpoints.

Applications.

Networks.

Databases.

Cloud services.

AI-native businesses introduce another useful way of thinking.

Instead of only asking:

What assets do we have?

Security teams also need to ask:

How are those assets connected through actions?

Consider:

Employee → Endpoint → AI Agent → Agent 2 → API → Application → Database → Customer Record

That is an execution graph.

Every node matters.

Every connection matters.

Every permission matters.

Every action matters.

The attack surface is therefore not just a list of systems.

It is the set of possible paths through those systems.

Suppose an organization discovers a vulnerability on a workstation.

Traditional vulnerability management asks:

How severe is the vulnerability?

That remains important.

But in an AI-native environment, another question becomes relevant:

What can this endpoint reach?

If the device is isolated, the operational consequences may be limited.

If the device has access to:

• AI development systems

• Cloud credentials

• Internal APIs

• Sensitive databases

• Administrative applications

the same vulnerability may have greater architectural significance.

This does not mean vulnerability severity changes.

It means organizations may need additional context when determining remediation priorities.

The relationship between assets becomes increasingly important.

Explore Vulnerability Management →

AI-native environments may produce enormous amounts of activity.

Agents may make API requests.

Applications may exchange data.

Automations may execute tasks.

Users may approve actions.

Multiple agents may communicate.

Security monitoring therefore cannot simply depend on raw volume.

It needs meaningful signals.

For example:

Agent A normally accesses customer records.

Then suddenly:

Agent A begins accessing administrative configuration data.

That behavioral change may deserve investigation.

Another example:

Agent B normally calls one API.

Then:

Agent B starts invoking several unrelated administrative services.

Again, context matters.

Threat detection increasingly becomes about identifying activity that is inconsistent with expected behavior.

A detection is still a signal requiring investigation.

It is not automatically proof of a confirmed incident.

In a traditional environment, security monitoring may be viewed primarily as an IT function.

In an AI-native organization, operational activity and security activity become increasingly interconnected.

If an agent performs a business action, the organization may need to know:

• What initiated it

• Which identity performed it

• Which agent was involved

• What data was accessed

• Which tools were used

• What application changed

• Whether the action was expected

• Whether additional actions followed

This makes security monitoring important not only for security teams but also for operational accountability.

The question becomes:

Can the organization explain what its software systems are doing?

One of the biggest differences between an AI-native business and a traditional business could be the amount of machine-to-machine activity.

Today, many business processes still involve humans moving information between systems.

An AI-native workflow may reduce some of those transitions.

For example:

CRM → Agent → Analytics → Proposal System → Billing → Notification

The systems may communicate without a human manually transferring information.

This can improve efficiency.

But it also increases the importance of machine identity and authorization.

When machines are performing more operations, organizations need stronger answers to:

Which machine performed this?

Which agent initiated it?

Who authorized the agent?

What policy allowed it?

There is an important distinction between human control and human execution.

A human does not necessarily need to perform every step.

Instead, humans may define:

• Policies

• Boundaries

• Approval requirements

• Business rules

• Risk thresholds

• Exceptions

AI systems can then perform permitted operations inside those boundaries.

This is similar to traditional access control.

A user does not need to ask an administrator for permission every time they open an authorized document.

The permission system establishes what they can do.

AI-native systems can follow the same principle.

The challenge is defining those boundaries correctly.

Traditional access policies may say:

User X can read system Y.

AI-native workflows may require more context:

Agent X can perform action Y when operating on behalf of user Z, against resource A, within workflow B, using approved tool C, subject to policy D.

That is a much richer authorization model.

It may consider:

• Identity

• Resource

• Action

• Context

• Risk

• Time

• Workflow

• Delegation

This does not mean every organization needs extremely complicated policies immediately.

It means that as AI becomes more autonomous, authorization models may need to become more contextual.

Imagine that an AI-native company experiences an unexpected financial transaction.

The organization should ideally be able to reconstruct:

1. Which customer interaction started the process

2. Which human was involved

3. Which endpoint was used

4. Which agent received the request

5. Which other agents participated

6. Which data was retrieved

7. Which tools were invoked

8. Which authorization was applied

9. Which application performed the transaction

10. What happened afterward

This is the difference between:

Something happened

and

We understand what happened.

Security teams need the second.

The phrase "AI-native" can create the impression that humans become irrelevant.

That is not what it means.

An AI-native business can still be fundamentally human.

People can remain responsible for:

• Strategy

• Governance

• Ethics

• Customer relationships

• High-impact decisions

• Risk management

• Security

• Accountability

AI may perform more operational work.

The organization still needs humans who define what the system is allowed to do.

The most important change may therefore be the distribution of work between people and software.

Imagine a company in 2028.

A customer submits a request.

An intake agent receives it.

A research agent gathers relevant information.

An analysis agent evaluates the request.

A policy agent checks organizational rules.

An operations agent updates the appropriate system.

A communication agent prepares the response.

A monitoring system records the activity.

A human reviews exceptions.

The architecture could look like:

Customer → AI Interface → Coordinator Agent → Research / Analysis / Operations → Data / Policy / Business Systems → Security Monitoring → Human Oversight

This is only a possible architecture.

The important point is that cybersecurity is present throughout the workflow.

If an organization builds its business around AI from the beginning, security cannot be an afterthought.

The architecture should consider security before agents are deployed.

Questions should include:

Identity

How is each agent identified?

Authorization

What is each agent allowed to do?

Data

What information can it access?

Tools

Which external capabilities can it invoke?

Delegation

Can it ask other agents to act?

Monitoring

How is activity recorded?

Human oversight

Which actions require approval?

Failure handling

What happens if an agent behaves unexpectedly?

Isolation

Can one compromised agent affect unrelated systems?

Recovery

Can the organization stop or reverse important actions?

These questions should become architectural requirements rather than emergency responses.

This is perhaps the most important conclusion of the five-year journey.

Cybersecurity is often described as protecting technology.

But in an AI-native business, technology increasingly performs the business itself.

Therefore:

Protecting technology becomes protecting operations.

If an AI system manages customer interactions, compromising it can affect customers.

If an agent manages financial operations, compromising it can affect finances.

If an AI system coordinates software development, compromising it can affect the software supply chain.

If an agent manages infrastructure, compromising it can affect the entire environment.

The closer AI moves toward the operational core of a company, the more closely cybersecurity becomes tied to business continuity.

The security platform must therefore provide visibility across the environment.

At the foundation:

**Endpoint Inventory** — Know what devices exist.

**Endpoint Visibility** — Understand what is happening on those devices.

**Endpoint Monitoring** — Observe activity over time.

**Vulnerability Management** — Identify weaknesses that could expose the environment.

**Threat Detection** — Identify potentially suspicious activity.

**Security Monitoring** — Correlate security-relevant information.

**Security Alerts** — Bring important signals to human attention.

Above those foundations, organizations will increasingly need to understand:

Identity

Applications

AI Agents

Agent Relationships

Tools

APIs

Data

Actions

The security platform does not need to replace every specialized system.

Its value comes from helping organizations maintain visibility and control across an increasingly interconnected environment.

The most important transformation between 2023 and 2028 may not be that AI became dramatically smarter.

It may be that AI became increasingly connected to the real world.

In 2023:

AI generated information.

In 2024:

AI accessed business context.

In 2025:

AI began taking actions.

In 2026:

AI became part of organizational workflows.

In 2027:

Multiple agents could potentially coordinate those workflows.

And in a possible 2028 scenario:

The business itself could be designed around AI-driven operations.

The security implications follow the same path.

The cybersecurity questions changed with the technology.

2023

What information are we giving AI?

The primary concern was data exposure and uncontrolled AI usage. See Part 1.

2024

What systems can AI access?

The concern expanded toward permissions, applications, APIs, identity, and context. See Part 2.

2025

What can AI do?

The focus shifted toward tools, actions, delegation, and agent authority. See Part 3.

2026

How do we secure an organization where AI can act?

The focus expanded to enterprise-wide visibility and control. See Part 4.

2027

How do we secure agents interacting with agents?

The focus expanded toward trust relationships, delegation, provenance, and cascading actions. See Part 5.

2028

How do we secure a business designed around AI?

The security architecture becomes part of the business architecture itself.

The future may be uncertain.

The preparation does not have to be.

Organizations can begin with fundamentals.

Know your endpoints

Maintain visibility into the devices used by employees.

Know your applications

Understand what software is operating across the organization.

Know your identities

Understand users, services, and emerging agent identities.

Know your data

Identify where sensitive information exists.

Know your AI usage

Understand which AI services employees and systems are using.

Know the permissions

Determine what AI systems can access.

Know the tools

Identify which agents can invoke external capabilities.

Monitor security activity

Maintain visibility into important events.

Manage vulnerabilities

Reduce weaknesses across endpoints and applications.

Investigate alerts

Treat security alerts as signals that require appropriate investigation.

Prepare for agentic workflows

Understand how AI may interact with existing systems before granting broad authority.

These steps are useful even if an organization never becomes fully AI-native.

The journey from 2023 to 2028 is not really a story about artificial intelligence replacing cybersecurity.

It is the opposite.

The more deeply AI becomes integrated into business operations, the more important cybersecurity becomes.

Generative AI created a new data boundary.

AI integrations created a new access boundary.

Agentic AI created a new action boundary.

Multi-agent systems created new trust boundaries.

An AI-native business creates a broader operational boundary.

The security architecture has to evolve with each one.

There is a temptation to describe the future simply as:

More AI.

That is too simplistic.

The more important transformation is:

More AI + More connectivity + More authority + More automation

Each additional capability increases the importance of control.

An AI system that can only answer questions has a limited operational footprint.

An AI system that can access company data has a larger footprint.

An AI system that can modify business systems has a larger footprint.

An AI system that can coordinate other agents has an even larger footprint.

An AI-native business may eventually depend on these systems for significant portions of its operations.

That is why cybersecurity cannot remain separate from AI strategy.

Organizations often compete through:

• Products

• Services

• Technology

• Data

• Customer experience

• Operational efficiency

In an AI-native environment, security becomes part of the foundation supporting all of these.

A company cannot confidently automate important processes if it cannot answer:

Who is acting?

What can they access?

What can they change?

What happened?

Can we stop it?

Can we recover?

The ability to answer these questions becomes an operational capability.

The five-year journey can be reduced to one principle:

As AI gains more ability to act, organizations need more visibility and control over the environment in which it acts.

That environment includes:

People

Endpoints

Identity

Applications

Agents

Tools

APIs

Data

Actions

Monitoring

Security cannot focus on only one component.

The system has to be understood as a whole.

The future of AI is not simply about creating systems that can generate better answers.

It is about creating systems that can participate in real work.

That work may involve customers, employees, financial systems, software, data, infrastructure, and business decisions.

As AI becomes more capable of performing that work, cybersecurity becomes increasingly connected to the organization's ability to operate safely.

An AI-native business therefore needs to think about security from the beginning.

Not because AI is inherently dangerous.

Not because every AI system will become autonomous.

And not because every organization will follow the same path.

But because authority, connectivity, data, identity, and action create security requirements regardless of whether the actor is human or machine.

The most important question for the next generation of businesses may therefore not be:

"How much AI can we deploy?"

It may be:

"How much AI can we securely trust with real work?"

That question brings the five-year journey back to where it started.

In 2023, we learned how powerful generative AI could be.

By 2028, the larger challenge may be determining how to build organizations where increasingly capable AI can operate without creating invisible security boundaries.

The answer will not be one model.

It will not be one firewall.

It will not be one endpoint agent.

It will not be one security product.

It will be an architecture.

An architecture built around:

Visibility.

Identity.

Least privilege.

Monitoring.

Threat detection.

Vulnerability management.

Human oversight.

Accountability.

Controlled autonomy.

And above all:

Security by design.

The journey is now complete:

2023 → AI generates

2024 → AI integrates

2025 → AI acts

2026 → AI operates

2027 → AI coordinates

2028 → AI becomes part of the business architecture

But there is one idea that remains constant throughout the entire journey:

Technology becomes more valuable when it can do more.

And technology becomes more important to secure when it can do more.

That is why the future of AI and the future of cybersecurity cannot be treated as separate conversations.

The future business will need both.

More capable AI.

More capable security.

And an architecture that allows the two to evolve together.

Return to the series introduction or continue with Part 4: The Agentic Enterprise.

Introduction: From Generative AI to AI Agents

Part 1: 2023: The Generative AI Revolution

Part 2: 2024: When AI Entered the Business Workflow

Part 3: 2025: The Rise of Agentic AI

Part 4: 2026: The Agentic Enterprise

Part 5: 2027: The Multi-Agent Organization (forward-looking scenario)

Part 6: 2028: The AI-Native Business and the New Definition of Cybersecurity (this article)