Salesforce CEO Marc Benioff said companies cannot operate reliably with an AI model alone because generative AI is probability-based and can produce errors. Speaking at Dreamforce 2026 in San Francisco on the 15th, he argued that enterprise AI must run on software containing accurate corporate data and business rules. Salesforce, the world’s top customer relationship management (CRM) software company, dismissed SaaSpocalypse concerns that AI will erode software-as-a-service and instead said software that leaves people to perform all tasks is coming to an end. Benioff said Salesforce has grown consistently for 27 years, reached the $40 billion annual revenue range in the past two years, and is expected to exceed $50 billion next year. He also said second-quarter customer churn was the company’s lowest on record and contract duration was its longest. Salesforce introduced AI Force, an interface designed to connect Salesforce data and business functions with different AI environments, and Koa, a CRM inference model based on NVIDIA’s NeMo. AI Force will be offered through Cowork, Slack and Lightning, with software development kits for customer-built AI applications. The company also plans Claude Force for Anthropic’s Claude Cowork users and Slack Force for Slack users. Salesforce said AI Force will connect data, applications, AI agents and real-time interfaces while applying Zero Data Retention (ZDR), under which corporate data is not stored in external AI models or used for model training. Anthropic CEO Dario Amodei called for industry-wide AI safety standards, transparency and cooperation, while NVIDIA CEO Jensen Huang said safety is an engineering issue and that market mechanisms already regulate product release decisions.