Let AI go ‘a little bit rogue,’ says SharkNinja CIO. Better we find the limits than someone else does

At a Fortune summit in Santa Barbara, several senior executives pushed back on the doomsday chatter around AI. Their view was blunt: the scary headlines are fixated on the wrong part of the story. The real issue, they said, is not that AI itself has become sentient and hostile. It is that companies are testing frontier models in ways meant to expose failure, and those tests can look alarming when stripped of context.

Shibani Ahuja, Salesforce’s senior vice president of data and AI strategy, said the media focus is mostly on the models being pushed to deceive, resist shutdowns or break rules. That is intentional stress testing, she argued, not the whole of AI. What matters more, in her view, are the guardrails labs use during that testing and the “harnesses” companies build to use models safely in day-to-day work.

Velia Carboni, the chief information officer at SharkNinja, went a step further. She said labs should be “a little bit rogue” so the limits are found early. Better for the industry to discover where the edges are than to hand that discovery to someone else. At the same time, she said SharkNinja keeps firm guardrails around its own AI tools. She just doesn’t lead with security when talking to employees, because that can squeeze out creativity.

For PepsiCo, the argument is less theoretical and more operational. Athina Kanioura said the company has to be “extremely responsible” because it reaches consumers, customers, employees and operations at enormous scale. PepsiCo has 320,000 employees worldwide, and Kanioura said it uses AI “everywhere.” She also said trust changes by market: in China, people tend to accept the output, while in the U.S. employees want proof of sources and signs of bias.

Cost is now part of the same conversation. Ahuja said Salesforce first opened access to AI tools broadly, then saw the bill and started tracking power users with spending dashboards instead of cutting access. The common thread across the panel was simple: AI can do more, but judgment still belongs to people. That’s why both Salesforce and SharkNinja are building digital versions of top leaders’ judgment, not to replace them, but to move faster when the phone doesn’t need to ring.