When the cloud first emerged as a viable option for enterprise IT, it exposed a fundamental question that needed to be answered: who is responsible when something goes wrong?
The answer to that question was not obvious at first, but after some hand-wringing and debate, the industry settled on the concept of shared responsibility. In that approach, neither the enterprise nor the hyperscaler holds full responsibility; each party is responsible for the attributes under its control, which means customers are expected to follow proper procedures to secure log-in credentials while the cloud provider promises to patch operating-system level software vulnerabilities promptly.
Vendors and standards groups are now adapting the shared responsibility concept for AI agents. However, the first point of confusion arrived quickly: Who answers for what an agent does?
"The biggest myth is that if you build on a major model provider, that provider is responsible for what your agent does," Rishi Bhargava, co-founder of Descope, told The Stack.
