Salesforce Introduces Trusted Enterprise AI Harness

Ahead of its Dreamforce conference (September 15–17 in San Francisco), the tech company announced Trusted Enterprise AI Harness. The open, modular platform architecture combines information from CRM, ERP, contracts, analytics, policies, and past interactions. This eliminates the need for companies to develop and manage these functions separately for each AI application.

Modules

• Trusted Enterprise AI Harness integrates six functions—context, agency, actions, governance, security, and models—into a unified architecture. This lays the foundation for AI to understand business contexts, plan tasks, execute actions, and do so while adhering to applicable rules and security requirements.
• Trusted Context: Consolidates customer context, data, metadata, knowledge, real-time signals, past interactions, and process knowledge. This gives AI a holistic understanding of customers and the business based on centrally managed enterprise data.
• Trusted Agency: Supports AI agents in planning, reasoning, remembering, and coordinating complex tasks. In doing so, Trusted Agency combines flexible AI-driven decision-making processes with fixed rules when clear and reliable decisions are required.
• Trusted Action: Securely connects AI to applications, interfaces, workflows, and business processes so that it can execute tasks. For example, AI can independently verify whether an order can be fulfilled on the same day and securely initiate the necessary steps.
• Trusted Governance: Manages data, policies, and processes with transparent quality requirements and guidelines. In this way, Trusted Governance ensures that AI is based on reliable information and complies with applicable corporate policies.
• Trusted Security: Protects data and systems. It regulates who is permitted to access which information and what actions AI can perform during ongoing operations. This ensures that AI accesses only authorized information and performs only authorized actions.
• Trusted Models: Selects the appropriate model for each task based on accuracy, performance, cost, and business requirements, and integrates it securely. This allows companies to deploy models flexibly and replace them as technologies and requirements evolve.

The Foundation for Secure and Traceable AI Actions

A seemingly simple customer question such as “Can this order be delivered today?” usually cannot be answered based on data from a single system: The CRM knows the customer and the relationship, the ERP system tracks inventory and delivery availability, and contracts contain commitments and entitlements. Analyses, policies, and past interactions – including purchase history and service requests – provide further critical information. The challenge, therefore, is not merely to connect these systems. AI must understand the entire context, determine the correct next action, execute it securely, and comply with corporate policies in the process. Trusted Enterprise AI Harness combines the capabilities of AI with the necessary business rules, security requirements, and control mechanisms – and makes them available company-wide for various agents and models.

New Centralized Control for AI in the Enterprise

With AI Control Plane, Salesforce creates a central location where companies can control and manage their AI. They can define identities and policies, manage AI deployment, evaluate its performance, behavior, and results, and monitor and control the costs of Salesforce and third-party AI. This gives companies a holistic view of their AI as it is deployed across workgroups, applications, models, and systems.

Open and Flexible for Existing Systems

Trusted Enterprise AI Harness can be integrated into existing IT landscapes and various work environments. Through MCP, APIs, skills, and plug-ins, its features can also be used outside of traditional Salesforce applications, including in Claude, Slack, Microsoft Teams, Agentforce, and other AI applications. In this way, Trusted Enterprise AI Harness supports the overarching AIforce strategy, which makes Salesforce available for any AI, agent, application, or interface. For existing Salesforce customers, the technologies they already use form the foundation for gradually expanding Trusted Enterprise AI Harness. These include Data 360, Informatica, MuleSoft, Agent Fabric, Tableau, Agentforce, Salesforce Guardian, and the Salesforce platform. This allows companies to gradually enhance their existing systems and solutions with new AI capabilities. Companies can deploy the six features as a package or as individual building blocks.