What experts think about Agentforce’s data privacy

Like other tech companies, Salesforce thinks AI agents could boost customers’ efficiency. However, the technology is very much in its early days, and customers are mixed about its security and whether it can carry out key tasks.

Salesforce executives Sabastian Niles, the company’s president and chief legal officer, and Paula Goldman, its chief ethical and humane use officer, told IT Brew that enterprises of all sizes could potentially benefit from agentic solutions. However, companies will need to trust that agentic AI can do the job, and that might come down to transparency and reliability.

Goldman said Salesforce is engaged in “robust” testing to ensure agentic AI behaves correctly.

“We’re continually monitoring and iterating, and sometimes that iteration is not necessarily about whether the technology has given the right answer,” Goldman said. “Sometimes it’s about when it’s the right moment for a person to step in and what the customer prefers.”

Defining the approach. Some industry experts, like TitanX Head of Marketing Evan Dunn, see a gap in customers’ understanding of what agentic AI truly means across the market. Salesforce’s customers and potential customers are “still waiting to watch something like Agentforce really come to the table and prove results in a way that feels like it matches the hype,” he said.

Salesforce defines agentic AI as the tech that powers AI agents so they’re able to “act autonomously without human oversight,” according to the company’s resource page. It claims agentic AI is more flexible and adaptable than “traditional AI systems that may struggle with complex and multistep tasks.”

Other tech giants also define agentic AI as a system that can perform complicated tasks without constant human oversight. For example, tech giants like IBM state agentic AI “can accomplish a specific goal with limited supervision,” while AWS views the tech as an “autonomous AI system that can act independently to achieve predetermined goals.”

And the data privacy approach. Dunn said “it’s really up to Salesforce to lose this opportunity” if it can’t ensure things like data quality or transparency.

For example, does an enterprise’s data stay in Salesforce’s control as it performs LLM-related activities, or is any portion of it routed through a third-party vendor (such as another commercial LLM)?

“I think there’s plenty of comms [Salesforce] could do around security, how agentic systems can have truly secure environments,” Dunn said. “I think a lot of us just don’t even know what that would look like or means, but we’re all eager to be educated. If they had subject matter experts get really loud about how they thought through this [in] all the different kinds of ways, I think that would go a long way.”

When asked about how Salesforce is prioritizing data privacy, safety, and trust, Niles pointed at giving customers tools to control and secure the agents’ use of data.

The buck stops there. Brian Soby, CTO and co-founder of AppOmni, a computer and network security company, told IT Brew that professionals should consider doing a meaningful security evaluation of Agentforce components, especially following the Salesloft Drift breach.

Additionally, he recommended that IT pros understand the difference between AI agents’ real capabilities and AI vendors’ often hyperbolic advertising.

Agentforce doesn’t do “crazy things out of the box,” Soby said. If an agent has the ability to do something, “it’s because the customer gave them the ability to do it.”

“Salesforce is and has always been a very powerful platform,” Soby said. “That’s a double-edged sword because if you say these agents are now at the crux of all of the business logic which makes them so powerful, you as an organization need to understand…what did you give this agent the ability to do? Because it can be almost anything on the platform, because they want to make the agents powerful and capable.”