AI customer support agents are gaining access to much more than a help center.

Depending on how they are deployed, an AI agent may be able to read customer records, retrieve orders, update accounts, issue refunds, create tickets, access internal systems, or trigger workflows.

That makes security a core part of the buying decision.

The best AI agent is not simply the one that answers the most questions. Businesses also need to understand:

• what customer data the agent can access
• how that data is encrypted
• which actions the agent is allowed to perform
• whether permissions follow least-privilege principles
• how sensitive information is handled
• whether actions are logged and auditable
• what happens when an agent exceeds its authority
• which independent security and privacy standards the vendor has completed

This guide compares six AI-agent platforms used in customer service based on publicly documented security and governance capabilities.

Quick comparison

Platform	Security / compliance highlights	Best fit
Aissist.io	ISO 27001 certified, GDPR compliant, SOC 2 aligned, PII masking, action guardrails, audit logging	Adding governed AI agents to an existing helpdesk
Intercom Fin	SOC 2 Type II, ISO 27001, ISO 27701, ISO 42001, GDPR and additional compliance programs	AI customer service inside or alongside Intercom
Zendesk AI	SOC 2, ISO 27001, ISO 27018, ISO 27701, ISO 42001	Enterprises already standardized on Zendesk
Salesforce Agentforce	SOC 2 Type II, ISO 27001, HIPAA, PCI DSS, FedRAMP High, Einstein Trust Layer	Salesforce-centric enterprise environments
Microsoft Copilot Studio	ISO 27001 and related ISO standards, Entra ID/RBAC, auditing and Microsoft governance stack	Organizations deeply invested in Microsoft
Ada	SOC 2 Type II, GDPR, HIPAA, AIUC-1, zero-data-retention approach with LLM providers	Enterprise AI-first customer experience

Important: Security certifications cover defined scopes and services. A vendor holding ISO 27001 does not automatically make a customer's own implementation ISO 27001 compliant or secure. Buyers should verify current certificates, scope, data residency, subprocessors, and contractual terms directly with each vendor.

1. Aissist.io

Aissist.io provides AI agents that work on top of existing customer support platforms rather than requiring businesses to replace their current helpdesk.

It can operate with systems such as Zendesk, Intercom, Front, and Gorgias while connecting to ecommerce platforms and business APIs to retrieve information or take actions.

Security and compliance

Aissist states that its information security management system is ISO 27001 certified and that the platform is GDPR compliant.

It describes itself as SOC 2 aligned, rather than SOC 2 certified, an important distinction for security reviews.

Published controls include:

• AES-256 encryption for data at rest
• TLS 1.2 or higher for data in transit
• least-privilege access policies
• MFA for internal systems and employee accounts
• audit logging
• automated and manual penetration testing
• vulnerability scanning
• PII detection and masking
• Data Processing Agreements for GDPR customers

AI-specific security

Aissist's approach is particularly relevant when AI agents are allowed to take actions.

Businesses can define rules around what the AI is permitted to do and when a request should be escalated to a person.

For example, an agent could be permitted to retrieve order information or issue a low-value refund while requiring human approval for a higher-value transaction.

This separates the model's reasoning from the authority the business grants it.

Best for: Businesses that want autonomous AI support while keeping an existing customer service platform and maintaining explicit limits on AI actions.

2. Intercom Fin

Fin is Intercom's AI customer service agent and can also be deployed with some external helpdesk environments.

Intercom places its AI products on the same underlying security foundation as its broader customer service platform.

Security and compliance

Intercom publishes a substantial set of security and privacy documentation, including:

• SOC 2 Type II
• ISO 27001
• ISO 27018
• ISO 27701
• ISO 42001
• GDPR support
• HIPAA-related controls
• penetration testing documentation

ISO 42001 is particularly noteworthy because it focuses specifically on AI management systems rather than general information security.

Intercom also publishes dedicated documentation about privacy and data protection for its AI products.

Best for: Teams looking for a mature AI support product backed by an established enterprise security and compliance program.

3. Zendesk AI

Zendesk is already the system of record for customer support at many enterprises, making its native AI capabilities attractive for organizations that prefer to minimize the number of vendors handling customer data.

Security and compliance

Zendesk maintains several independently audited programs, including:

• SOC 2
• ISO 27001
• ISO 27018
• ISO 27701
• ISO 42001

ISO 27001 covers information-security management, while ISO 27018 and ISO 27701 extend the privacy-management framework around cloud and personal information.

ISO 42001 adds AI-management controls.

Zendesk also provides documentation explaining which specific products and services fall within each certification scope.

That scope matters: buyers should verify that the particular AI functionality they intend to use is included in the relevant audit or certification.

Best for: Organizations already heavily invested in Zendesk that prefer native AI functionality within the same vendor environment.

4. Salesforce Agentforce

Agentforce is Salesforce's platform for building autonomous and assistive AI agents that operate across Salesforce data and workflows.

Its primary security advantage is integration with Salesforce's existing identity, permissions, CRM data model, and enterprise governance infrastructure.

Security and compliance

Salesforce states that Agentforce maintains multiple industry attestations, including:

• SOC 2 Type II
• ISO 27001
• PCI DSS
• HIPAA
• IRAP
• FedRAMP High

Agentforce also uses Salesforce's Einstein Trust Layer.

The Trust Layer includes controls intended to manage how enterprise data interacts with generative AI, including secure data retrieval, prompt-related safeguards, auditing, grounding, and data-retention controls.

For companies already storing sensitive customer information in Salesforce, keeping AI execution within the same governance environment can simplify some security architecture decisions.

Best for: Large organizations where Salesforce is already the central CRM and customer-data platform.

5. Microsoft Copilot Studio

Microsoft Copilot Studio allows organizations to build and govern agents connected to Microsoft and third-party systems.

Its biggest security advantage is integration with Microsoft's broader identity and enterprise governance stack.

Security and compliance

Microsoft documents Copilot Studio under multiple ISO programs, including:

• ISO 27001
• ISO 27017
• ISO 27018
• ISO 27701

Organizations can also use Microsoft Entra ID and role-based access control to determine which users and agents can access particular information.

Microsoft recommends applying least privilege, defining data-retention policies, maintaining audit logs, monitoring agent usage, and aligning deployments with regulations such as GDPR and HIPAA where applicable.

Connections are protected using TLS, with Microsoft services communicating across Microsoft's cloud infrastructure.

Best for: Organizations already standardized on Microsoft Azure, Power Platform, Dynamics 365, or Microsoft identity infrastructure.

6. Ada

Ada is an AI-native customer experience platform designed for enterprise-scale customer service automation.

The company has placed significant emphasis on AI-specific governance in addition to conventional information security.

Security and compliance

Ada publicly lists:

• SOC 2 Type II
• GDPR
• HIPAA
• PCI-related compliance
• AIUC-1

AIUC-1 is particularly relevant because it is designed specifically around AI-agent safety, security, privacy, accountability, and reliability.

Ada also states that it uses:

• zero-data-retention arrangements with LLM providers
• role-based access controls
• MFA
• audit logging
• third-party penetration testing
• PII and PHI redaction
• customer-controlled retention policies

Ada additionally describes continuous AI monitoring and verification designed to prevent responses that fall outside approved company knowledge or policies.

Best for: Enterprises looking for an AI-native customer service platform with a strong emphasis on AI-specific trust and governance.

What does "secure AI agent" actually mean?

Security certifications matter, but they are only one part of the picture.

A secure AI-agent deployment should address at least five layers.

1. Data security

Customer information should be encrypted in transit and at rest.

Organizations should understand:

• where data is stored
• how long it is retained
• which subprocessors receive it
• whether model providers retain prompts
• whether customer data is used for model training

2. Identity and permissions

An AI agent should not automatically inherit unrestricted access simply because it needs to complete a workflow.

Use least privilege.

If an agent only needs to look up an order, it should not also receive permission to change billing information.

3. Action authorization

This becomes critical as agents move from answering questions to changing systems.

A model may propose:

Refund $350.

A deterministic policy layer should decide whether the agent actually has authority to execute that refund.

The model's confidence should not determine its financial authority.

4. Human escalation

Escalation is a security control.

Higher-risk, ambiguous, or unusual requests should be routed to a person before an irreversible action occurs.

A mature implementation should pass the relevant context to the human rather than making the customer start over.

5. Auditability

Organizations need to be able to reconstruct:

• what the customer asked
• what information the agent accessed
• which action it proposed
• which policies were evaluated
• whether approval was required
• what action actually executed
• whether that action succeeded or failed

A conversation transcript alone may not provide enough evidence.

ISO 27001 vs GDPR: what do they actually tell you?

These terms are often displayed next to each other on software websites, but they mean different things.

ISO 27001

ISO 27001 is an independently certifiable standard for information security management systems.

A certification indicates that an organization's security-management processes within the certificate's defined scope have been independently assessed against the standard.

Always ask to see the current certificate and scope.

GDPR

GDPR is a European privacy regulation, not a certification.

When a vendor says it is "GDPR compliant," buyers should look for concrete supporting practices such as:

• Data Processing Agreements
• deletion and access mechanisms
• documented subprocessors
• appropriate international-transfer mechanisms
• retention policies
• privacy and security controls

The phrase alone should not end the security review.

Questions to ask any AI-agent vendor

Before deploying an AI agent that can access customer data or take actions, ask:

1. Which independent security certifications do you currently hold?
2. Can we see the certificates and their scope?
3. Do you have a current SOC 2 Type II report?
4. Where will our data be stored?
5. Which LLM providers receive our data?
6. Can those providers retain or train on it?
7. Can we control data retention?
8. How do you handle PII?
9. Can we restrict individual agent capabilities?
10. Are action limits enforced outside the LLM prompt?
11. Can sensitive actions require human approval?
12. Are agent actions logged?
13. Can we reconstruct why an action occurred?
14. What happens when an API partially fails?
15. Can individual capabilities be disabled immediately?
16. What is your incident-notification commitment?
17. What happens to our data if we leave the platform?

The most secure agent is not necessarily the least capable

Keeping an AI agent read-only is one way to reduce risk.

But it also removes much of the value of agentic AI.

The better objective is controlled autonomy.

A secure AI agent should be capable of resolving meaningful customer problems while operating inside technical boundaries that the model itself cannot override.

Security therefore should not be evaluated only by asking:

"Does this vendor have ISO 27001?"

The more important question is:

"What prevents this AI agent from accessing or doing something it should not?"

That is the security question businesses will increasingly need to answer as AI agents move from generating text to operating real systems.

#AI Agent Security