Artificial intelligence is making Salesforce more powerful, but it is also raising an uncomfortable question for businesses: just because AI can access your customer data, does that mean it should?
As organizations introduce AI agents into sales, service and other customer-facing processes, the information stored inside Salesforce is becoming more valuable than ever. Customer histories, financial details, internal notes and sensitive personal information can all help AI deliver more relevant responses and take meaningful action.
But that same access creates risk. Data that was previously viewed by a small group of employees could potentially become available through new AI-powered workflows if permissions, integrations and security controls are not properly designed.
For Salesforce leaders, the challenge is no longer simply making data available to AI. It is understanding which information AI should use, how it should be protected and who remains accountable when something goes wrong.
Traditional CRM security largely focuses on controlling which users can access particular records, fields and systems. AI introduces another layer of complexity because agents may retrieve information, interpret it and use it to generate responses or perform tasks.
Consider a customer service agent using Salesforce AI to resolve an account query. It might need access to purchase history, previous interactions and relevant support records. But should it also be able to retrieve confidential internal notes, sensitive financial information or details about unrelated customers?
The answer depends on the task, the organization's policies and the permissions governing the agent. Simply connecting AI to a CRM does not automatically mean every record becomes accessible. However, poorly configured access or overly broad integrations can create opportunities for information to be exposed or used inappropriately
This makes data security an essential part of Salesforce AI planning, particularly for organizations operating in regulated industries or handling large volumes of personal information.
Most organizations know they hold sensitive customer data.
Few, however, have a complete picture of where that information sits, how it moves between systems and whether it is consistently protected.
Over time, Salesforce environments can accumulate information across standard fields, custom objects, attachments, case histories and integrations. Employees may also enter sensitive details into free-text fields that were never designed to store them.
Potential problem areas include:
The challenge becomes more complicated when information from multiple sources is brought together to support AI-powered experiences.
For example, a field that appears harmless on its own may become sensitive when combined with other customer information. An AI system capable of connecting those details can reveal relationships that were previously difficult to identify.
Before expanding AI access, organizations need to understand what information they hold, where it originates and whether existing classifications and permissions reflect its actual sensitivity.
Salesforce AI can potentially access sensitive information when that information is available through the data sources, tools and permissions configured for a particular AI experience. The precise access depends on the Salesforce products involved, the organization's security configuration and how the AI workflow has been implemented.
That distinction matters. AI access is not necessarily unrestricted, but it should never be assumed to be safe simply because the underlying platform includes security features.
Salesforce leaders should examine how AI agents authenticate, which records and fields they can retrieve, what external systems they interact with and whether their actions are appropriately logged and monitored.
An agent that only needs to answer general product questions should not automatically receive the same access as an employee managing sensitive customer accounts.
The principle is familiar: give systems the minimum access required to complete their intended tasks. Applying it consistently becomes increasingly important as AI agents take on more responsibilities.
Recent Salesforce developments highlight why organizations need a more complete approach to protecting customer information.
In October 2026, Salesforce published new guidance on sensitive data discovery in Data 360, highlighting how organizations can identify sensitive information before it reaches AI models. Salesforce has also emphasized the importance of protecting information throughout its lifecycle, including backups and archives.
These capabilities address different parts of the same problem. Data discovery helps businesses understand what needs protecting, while encryption helps protect information from unauthorized exposure. Neither removes the need for carefully designed access controls, retention policies and monitoring.
Think of it like securing a building. Encryption helps protect what's inside, but organizations still need to know which rooms contain valuable information, who holds the keys and whether anyone is entering places they shouldn't.
For Salesforce AI, that means treating data discovery, permissions and encryption as connected responsibilities rather than separate technical projects.
Businesses do not need to solve every data challenge before exploring AI. They do, however, need to understand the risks associated with the specific workflows they intend to automate.
These five questions provide a practical starting point.
Start by identifying personal, financial, confidential and regulated information across Salesforce and connected systems. Pay particular attention to free-text fields, attachments and historical records that may not have been classified consistently.
Review the permissions associated with agents, their underlying actions and any connected applications. Access should reflect the task being performed, rather than simply inheriting broad permissions because they are easier to configure.
Agents may receive incomplete, misleading or malicious instructions through customer messages and retrieved content. Testing should examine whether an agent can be manipulated into revealing information or taking actions outside its intended purpose.
Organizations need appropriate visibility into agent activity, including information accessed, actions performed and exceptions encountered. Audit records should support investigation and accountability without unnecessarily exposing additional sensitive information.
AI governance requires clear ownership. Security teams can define controls, Salesforce specialists can implement them and business leaders can establish acceptable behavior. But responsibility must remain clear when agents interact with real customers and business processes.
Salesforce's shared responsibility approach reinforces this distinction: Salesforce provides foundational security capabilities, while organizations remain responsible for configuring permissions, access and appropriate guardrails.
As AI becomes embedded in business operations, organizations increasingly need expertise that extends beyond traditional CRM configuration.
A secure Salesforce AI implementation may involve Salesforce Architects, data specialists, security professionals, AI Engineers and people who understand the business processes being automated. These disciplines need to work together from the beginning, particularly when AI agents interact with external systems or take actions on behalf of employees.
This is where the distinction between implementing a feature and engineering a reliable AI solution becomes important.
For example, an organization might want an AI agent to retrieve customer information from Salesforce, check details against another enterprise system and initiate a follow-up action. Building that workflow requires decisions about integration architecture, data access, exception handling, monitoring and human oversight.
Through Tenth Revolution Group's wider Talent, Training and Transformation capabilities, organizations can access specialist Salesforce expertise alongside broader Applied AI engineering support. That includes Forward Deployed Engineers who work closely with business and technical teams to build AI solutions around real operational requirements.
The role of these engineers is particularly relevant when organizations need to connect AI capabilities to existing systems, test them in real-world conditions and refine how they operate.
For our experts, the wider lesson is that Salesforce AI projects increasingly require coordinated expertise across multiple disciplines. Finding the right people, developing existing teams and bringing in specialist engineering support can all form part of the solution.
The next phase of Salesforce AI will create opportunities to improve customer service, reduce manual work and make better use of business information. But those benefits depend on organizations understanding the data their AI systems rely on.
Salesforce's latest focus on sensitive data discovery is a useful reminder that AI readiness involves more than connecting information and deploying agents. Businesses need to know what data they hold, which systems and people can access it and how those controls behave when AI enters the picture.
The organizations that make the most progress will be those that treat security as part of designing useful AI, rather than something to revisit after deployment.
Because when an AI agent can act on your customer data, understanding what it is allowed to do becomes just as important as understanding what it is capable of doing.
Whether you're strengthening Salesforce security, preparing customer data for AI or introducing more advanced workflows, having the right expertise makes a difference.
Powered by Tenth Revolution Group, Mason Frank helps organizations access specialist Salesforce professionals across architecture, development, data and platform management. We can also support broader Applied AI engineering, training and transformation requirements.
Tenth Revolution Group helps organizations access the expertise, develop the skills and deliver the solutions needed to make Salesforce work in real business environments.