What one privacy expert thinks about challenges facing agentic AI

What would it take for everyone to completely trust AI’s latest commercial models? That’s not a rhetorical question—we’re really asking.

Ed Britan, SVP of global privacy at Salesforce, sat down with IT Brew during Agentforce World Tour in Washington, DC, to discuss the challenges surrounding consumer privacy and trust in the context of agentic AI and data collection.

During the discussion, Britan pointed to how public commentary can potentially influence consumer opinions of enterprise AI. (For example, just last month, Anthropic CEO Dario Amodei said it’s hard to know whether or not an AI model is truly conscious.) Our chat also covered privacy, AI innovation in the B2B space, and more.

This interview has been edited for length and clarity.

When we’re talking about agentic, what are some common challenges you’re seeing? I know there’s this lack of trust, but how do you walk the balance of understanding risk and helping customers innovate?

I want to make clear the distinction for customers, because they see in the press, some other company’s head of AI is talking about…AI’s helping customers, helping individuals work through emotional situations, and they say this because they want to sound really cutting, like it’s amazing and scary.

In the press the other day, the Anthropic CEO—I think he said this just because it sounds really powerful—but that his AI has feelings. I think that sounds [like] that’s really advanced, and Anthropic’s tech is at the cutting-edge, but then I think those kinds of things make our customers very nervous because our customers are pure B2B. They don’t want the AI to be an emotional support tool for their employees, they want to trust their business partner.

We’re dealing with really valuable business data that customers are entrusting us with…they need agents they can trust from a business perspective, with the same enterprise-grade commitment to mentality.

That stuff drives me crazy, personally, too, just because I feel like it hurts—I have to combat that stuff whenever those kinds of stories come out because the customers see it. Tech isn’t a monolith, but I think people often view tech as a monolith. When one part of tech does something, they view it as all tech. But there’s a big difference on the business model, the type of company you are.

I would love to know more about that, it sounds like there’s this generalist, sweeping language that’s bothersome.

Yeah, and we’re platforms, we do offer customers the option of LLMs, and we don’t want to dictate for them. Which, maybe they want to use Claude because it’s the best, lots of customers think it’s the best LLM out there, and so our goal is to provide optionality. And there is enterprise-grade Claude, and…they all offer B2B services as well, and I trust that they’re meeting their enterprise-grade commitments with those B2B services.

The consumer space is very different, where companies have direct control over the engagement with the individual, whereas we give all the control to the customer.

Everyone’s talking about agentic, but what is on the horizon? What’s next in the AI story?

We need to really solidify the agentic, I feel like we haven’t even come close. Even just on my time, I’m trying to increase the number of agents more, it’s making our lives incredibly easier.

There’s going to be more agent-to-agent conversation. Right now, it’s very much human-to-agent…agents are always gonna be controlled by humans, but [it’s going to be] increasingly agent-to-agent activity. I think that’s actually exciting too, because it’s going to automate more, it’ll just create more efficiencies, more scale.

The standards for that haven’t been set yet, and I’m a lawyer, so I’m not going to build the tech but I can help build the standards. I think we need to have global standards around this. I think you can create those global standards through external engagement…I’m excited about setting the standards for what agents should be, and what agent-to-agent negotiation should be, and ensuring that everything’s really strong from a privacy perspective.