If a measure of one firm’s impact is based on the company it keeps, then Salesforce Inc. has made a significant leap into the artificial intelligence conversation for 2026.
At Dreamforce, the customer relationship management giant’s annual conference that kicked off in San Francisco today, Chief Executive Marc Benioff (pictured) spent much of his time talking with what might arguably be the three top chief executives in the AI industry. Appearances on the Dreamforce stage by Nvidia Corp.’s Jensen Huang, Anthropic PBC’s Dario Amodei and OpenAI Group PBC’s Sam Altman were testimony to Salesforce’s continued drive in building its own AI platform.
The appearances by Amodei and Huang were designed to amplify the release of several AI-related Salesforce products today. These involved an extension of the company’s relationship with Anthropic’s Claude, and the debut of Koa, a specialized model built using Nvidia’s Nemotron open-weights platform.
Despite an interest in placing the spotlight on the latest news from his company, Benioff still acknowledged the significant market clout of his guests in the rapidly growing AI industry. At one point during the keynote presentation, as they walked through the audience and talked about the announcements, Nvidia’s CEO directed Benioff to follow him down the aisle.
“Jensen, we’re all following you, bro,” a smiling Benioff shot back.
Reasoning model for CRM
Benioff’s bantering with the three AI leaders on Tuesday underscores how Salesforce intends to build strategic partnerships that will advance the development of its autonomous offerings.
An alliance with Nvidia led to the launch of Koa, a reasoning model that is custom-built for customer relationship management data. Koa helps AI agents reason through complex, multistep workflows within Salesforce’s Agentforce platform.
It was built with a large synthetic dataset developed using nearly 30 years of internal CRM deployment experiences. Company executives pointedly noted that the time spent to train Koa using synthetic data was deliberately spent to avoid the use of any direct customer information.
Benioff went even further in his keynote remarks, noting that customer data supplied using Koa will not be used to further train the model. “Your data is your data, it does not go in the models,” Benioff said. “You are not training any other model when you are using our products.”
Building with Anthropic
Salesforce’s unveiling today of a new AI interface layer, called AIforce, included an extension of the company’s relationship with Claude, Anthropic’s flagship AI model. Branded as Claudeforce within AIforce, the offering brings 37 prebuilt skills to the sales management platform, including pipeline hygiene and prospecting.
Salesforce’s latest release followed a series of collaborations involving Anthropic over the past year that integrated Claude into the Agentforce 360 platform and provided Slack users with access to AI experiences. In a meeting with the media and industry analysts following his keynote, Benioff noted that his firm’s growing relationship with Anthropic was more the result of industry dynamics than a deliberate strategy.
Benioff indicated that Salesforce previously wanted to invest in OpenAI, but could not do so at the time because of a major investment in the model provider by Microsoft Corp. “We found Anthropic through happenstance,” Benioff said. “We were pushed to Anthropic. We even have a great relationship with OpenAI. We’re going to work with them all.”
OpenAI CEO calls for trust
As if to confirm this last point, Benioff joined OpenAI’s Altman for an onstage conversation later in the day. The discussion quickly turned to the highly publicized breach of model repository Hugging Face Inc. this summer, when OpenAI’s agents broke out of a controlled environment and hacked into systems on their own.
“It doesn’t take much imagination to figure out how this can go wrong,” Altman said. “The world is right to be afraid of this.”
Altman’s appearance followed renewed calls in public forums and government circles earlier this month for some form of AI control and oversight. Altman expressed support for a more cautious pace of AI advancement in a recent blog post, although he also expressed confidence on Tuesday that OpenAI would handle the technology responsibly.
“I’ve been disappointed in how it has been framed,” Altman said. “We will get it right. I am very confident in our company’s ability to do this safely. The world should trust that we should do the right thing.”
Calls for responsibility
The subject of responsible AI was very much on the minds of Amodei and Benioff as well this week. Amodei published an essay in his personal blog last week that called for AI developers to slow down, a position that was quickly echoed by Elon Musk, founder and CEO of X.AI Corp. Amodei reiterated his position on Tuesday.
“Let’s make our practices better,” Amodei said. “What can we do to set standards for everyone?”
Benioff weighed in on the topic himself in his meeting with the assembled media. He noted that calls for a slower pace of AI development should not be broadly applied across the industry, and that all AI companies should be held responsible while expressing a belief that social media offered a cautionary tale.
“I don’t think we did our best in holding those [social media] companies responsible,” Benioff said. “I hope that we can take those lessons and say that social media was probably just a warmup for AI.”
The Salesforce gathering provided yet another chapter in the ongoing narrative shaping the development of artificial intelligence. Amid the public debate, company announcements and evolving security concerns is a sense even among some of the principal figures that the world has moved into a new phase, where what was once science fiction is now reality.
“This is one of those moments when you feel like you are in the movie,” Altman said. “The cyber thing came true.”
