Claudeforce is the expanded partnership between Salesforce and Anthropic, announced on August 26, 2026, that puts the data, workflows and business rules held in Salesforce to work inside Claude. Nearly three weeks later, at Dreamforce on September 15, 2026, Salesforce folded Claudeforce into AIforce, an interface layer that brings the CRM to wherever people and AI agents work.
In this article from the artificial intelligence section, TOT summarizes the announcements, the data safeguards, the partnership timeline and what they mean for businesses in Vietnam.
>>> Read more:
What is Claudeforce
Claudeforce is the name of the expanded strategic partnership between Salesforce and Anthropic, announced on August 26, 2026. In the words of the press release, it brings “Claude’s intelligence and reasoning together with Salesforce’s trusted enterprise harness to make its data, workflows, business logic, actions, and governance securely accessible to power agentic experiences wherever work happens.”
According to the press release, Claudeforce covers three product areas:
- Salesforce in Claude: a plugin with 37 prebuilt sales skills, such as meeting prep and deal health review; an administrator sets it up once, centrally, so there is nothing for individual users to install.
- Claude in Salesforce and Agentforce: Claude is the reasoning model for the Atlas Reasoning Engine and the default model for Agentforce Vibes and Agentforce Coworker, running inside the Salesforce Trust Boundary via Amazon Bedrock.
- Claude in Slack: the default model for Slack AI and Slackbot.
The partnership runs both ways. Anthropic has chosen Salesforce as its preferred CRM. Salesforce, for its part, reports that running Slackbot on Claude internally has delivered 8.1 million hours of annualized productivity gains.
AIforce at Dreamforce 2026: Salesforce meets users where they work
AIforce is, in Salesforce’s words, “a live interface layer that brings the full power of Salesforce to wherever people and agents work,” with no need to open the traditional Salesforce application. Salesforce unveiled AIforce during the Dreamforce 2026 keynote in San Francisco on September 15, 2026; Claudeforce is one of its four components.
Marc Benioff, CEO of Salesforce, said: “AI is creating an interface revolution…combining model intelligence with all the context that customers have built into Salesforce.” AIforce has four components:
| Component | What it includes |
|---|---|
| Claudeforce | Salesforce in Claude: a prebuilt MCP (Model Context Protocol) server/plugin with 37 sales skills; skills for Tableau, service, marketing and commerce are to follow. Also includes the Salesforce Development plug-in for Claude Code (more than 40 skills). |
| Slackforce | Slackforce Surfaces (interfaces that pull context from Salesforce, Slack and other tools), Slackbot, Slack CRM (update records with a prompt) and Slack Code (a multi-user AI development environment). |
| Agentforce Coworker | An AI teammate inside the Lightning interface; according to Salesforce, 100,000 users activated it in the first 35 days. |
| Headless Toolkit | MCP, APIs, plugins, skills and tools for developers. |
The AIforce announcement of September 15 states that Salesforce in Claude is “available in beta to all customers”; Anthropic says the plugin is “available in beta on all paid Claude plans.” The August 26 press release had mentioned only select pilot customers, with an open beta expected in September 2026.
On the same day, September 15, Salesforce announced Koa, described as “Salesforce’s first CRM reasoning model,” built in-house for Agentforce. The Koa press release does not mention Claude; Koa is in pilot with select customers, with general availability (GA) planned for winter 2026 in U.S. regions. Salesforce is therefore taking a multi-model approach: Claude for Atlas and Agentforce, alongside its own Koa.
How Salesforce in Claude works
Salesforce in Claude works as a plugin: a seller makes a request in Claude, and the plugin reads and writes Salesforce data within that user’s Salesforce permissions. The connection runs over MCP, the open protocol that links AI models to tools and data. Salesforce remains the system of record.
According to the Claudeforce product page (accessed September 16, 2026), the 37 skills fall into five groups, which overlap with many common workflows for AI agents:
- Daily Workflow: daily briefings and pipeline reviews.
- Deal Strategy: close planning and objection handling.
- Account Growth: account planning and lead triage.
- Meeting Preparation: call prep and conversation summarization.
- Governance & Action: activity logging and win-loss analysis.
Three setup steps
According to the Claude Help Center:
- The Salesforce administrator signs up for the beta through AgentExchange and follows the setup guide.
- The Claude administrator (Primary Owner or Owner role) chooses the installation method under Organization settings > Plugins, configures the Salesforce MCP Connector with OAuth and selects which groups may use it.
- Members enable the plugin and sign in with their own Salesforce accounts.
Permissions and approvals
Claude sees only what each user’s existing Salesforce permissions already allow, so the business does not have to build a new permission model or migrate data. For write operations, Claude asks for approval by default: the user approves each proposed change before it is written to Salesforce.
Beta eligibility
Eligibility for the beta comes down to three areas, according to Anthropic and Salesforce:
- On the Salesforce side: the latest Sales Cloud Enterprise Edition and approval from Salesforce through the beta sign-up.
- On the Claude side: a paid Claude plan; the plugin works in chat and Claude Cowork (web and desktop).
- Scope: only the Sales skills are available for now; the other product areas are marked “Coming soon,” and additional skills will begin launching in late 2026. Pricing has not been announced; Salesforce notes that pricing is subject to change and availability varies by region.
How business data is protected
According to Salesforce and Anthropic, business data is protected by four layers: zero data retention, the Salesforce Trust Boundary, no training on customer data by default, and Enterprise Frontier Safeguards (EFS). These are commitments the two companies have published themselves; businesses should still check them against their contracts.
Zero data retention. The AIforce announcement is explicit: business data is used to answer the question at hand and is not retained by the model provider. According to Salesforce Ben, Benioff put it plainly: “Your data is your data. It does not go in the models.”
Trust Boundary and Amazon Bedrock. According to Anthropic’s announcement of October 14, 2025, the company is the first large language model (LLM) provider fully integrated within the Salesforce Trust Boundary via Amazon Bedrock, with all of Claude’s traffic contained within the Salesforce virtual private cloud.
No training by default. Anthropic states that on Team and Enterprise plans it does not train its models on customer data by default. It has also said that it has never trained on enterprise data without explicit permission, and never will.
Enterprise Frontier Safeguards (EFS). EFS is a set of privacy protections and abuse-detection measures that Anthropic is developing together with its customers, Salesforce among them. Data stays in customer-controlled cloud infrastructure, under the customer’s own encryption keys.
EFS will roll out in phases starting later in fall 2026; until then, eligible customers receive zero data retention on Claude Fable 5 and 5.1, according to Anthropic.
The Salesforce–Anthropic partnership timeline
Claudeforce and AIforce are the result of a partnership that dates back to 2024:
- 2024: Claude is integrated into Einstein and Agentforce.
- October 14, 2025: The two companies expand the partnership. Claude becomes a preferred model for Agentforce in regulated industries (financial services, healthcare, cybersecurity and life sciences); Claude is fully integrated within the Trust Boundary; Salesforce deploys Claude Code across its global engineering organization.
- February 2, 2026: Salesforce announces Claude in Slack (available), Agentforce 360 with Claude (coming soon) and MCP Apps.
- August 26, 2026: The two companies announce Claudeforce on the same day Salesforce reports its results for the second quarter of fiscal 2027.
- September 15, 2026: At Dreamforce 2026, Salesforce launches AIforce, moves Salesforce in Claude into beta and announces Koa; according to Bloomberg, 7,000 Salesforce employees have already used Claude-integrated tools.
What it means for businesses in Vietnam
For businesses in Vietnam, Claudeforce above all shows where CRM is heading: the sales team works inside an AI interface, while the CRM sits behind it as the system of record and the permission layer.
Businesses already on Sales Cloud
Eligible businesses can sign up for the beta through AgentExchange. Before switching it on for the sales team, a business should review its Salesforce permissions: loose access rights in the CRM carry over unchanged into the AI.
Businesses on another CRM or a custom-built one
The same approach applies: the CRM is brought into the AI interface through MCP and APIs, rather than forcing users to open the CRM application. A company’s own CRM software or ERP, given clean APIs and sound permissions, can also be connected to Claude or another AI assistant.
Data, approvals and sales-team training
A business needs to prepare three things:
- Decide which data may be exposed to the AI and write the security commitments into the contract.
- Keep a human approval step before the AI writes any data, as Salesforce in Claude does.
- Train the sales team to phrase requests and to check the results.
No announcement yet on Vietnamese support or regional availability
As of September 17, 2026, Salesforce and Anthropic have not announced supported regions or Vietnamese language support for Salesforce in Claude; Salesforce notes only that availability varies by region. Businesses should keep an eye on the Claudeforce page and AgentExchange.
TOT’s perspective
TOT advises on and builds CRM and ERP integrations with AI agents through APIs and MCP, following the same principles that Claudeforce applies: the business system is the system of record; the AI agent reads and writes only within its granted permissions; and every write goes through approval and is logged.
See more: how businesses build AI agents.
Frequently asked questions about Claudeforce
What is Claudeforce?
Claudeforce is the expanded strategic partnership between Salesforce and Anthropic, announced on August 26, 2026, that brings the data and business rules held in Salesforce into Claude. Claudeforce comprises Salesforce in Claude, Claude in Agentforce and Claude in Slack; since September 15, 2026, it has been one component of AIforce.
Is Salesforce in Claude free?
Pricing is not yet known: Salesforce has not announced it and notes that it is subject to change. To join the beta, an organization needs a paid Claude plan, the latest Sales Cloud Enterprise Edition and approval from Salesforce through the sign-up on AgentExchange.
Is Claude trained on a company’s Salesforce data?
Not by default: on Team and Enterprise plans, Anthropic does not use customer data to train its models. Salesforce adds a zero data retention commitment: data is used only for the question being processed and is not retained by the model provider.
How is AIforce different from Agentforce?
Agentforce is Salesforce’s AI agent platform; AIforce is the interface layer that brings Salesforce capabilities to wherever users and AI agents work. AIforce comprises Claudeforce, Slackforce, Agentforce Coworker and the Headless Toolkit; Agentforce Coworker is the part of Agentforce that has been brought into AIforce.
Can businesses in Vietnam use Salesforce in Claude yet?
There has been no specific announcement about Vietnam or the Vietnamese language. Anthropic lists only the general conditions: a paid Claude plan, the latest Sales Cloud Enterprise Edition and beta approval from Salesforce; Salesforce notes that availability varies by region. Businesses should sign up and wait for Salesforce to confirm.
Is Koa the same as Claude?
No. Koa is the first CRM reasoning model built by Salesforce itself for Agentforce, announced on September 15, 2026, in a separate press release that does not mention Claude. Claude remains the reasoning model for the Atlas Reasoning Engine and the default model for Agentforce Vibes, Agentforce Coworker, Slack AI and Slackbot.
Conclusion
Claudeforce shows Salesforce and Anthropic working on one problem: bringing the CRM to wherever users work while keeping permissions, business rules and data controls intact. AIforce extends that idea to Slack, Lightning and developer tools.
For businesses in Vietnam, the immediate value lies in the approach: the CRM or ERP is the system of record, AI agents connect through APIs and MCP, and every write goes through approval and is logged. Those three principles hold for Salesforce in Claude and for in-house systems alike.
Sources
- Salesforce press release on Claudeforce, August 26, 2026
- Claudeforce product page
- Salesforce’s AIforce announcement, September 15, 2026
- Anthropic blog post on Salesforce in Claude, September 15, 2026
- Claude Help Center: Set up Salesforce in Claude for your organization
- Anthropic introduces Enterprise Frontier Safeguards, September 1, 2026
- Anthropic expands its partnership with Salesforce, October 14, 2025
- Salesforce: trusted context and AI actions on Claude, February 2, 2026
- Bloomberg (republished by The Spokesman-Review) on Dreamforce, September 15, 2026
- Salesforce Ben’s roundup of the Dreamforce 2026 keynote, September 16, 2026
- Salesforce press release on Koa, September 15, 2026
