Salesforce Repurposes Hallucinating AI Agent to Audit Company Data

Salesforce, Inc. temporarily disabled a newly launched AI agent on its website after the tool produced inconsistent results and hallucinations. Shibani Ahuja, senior vice president of enterprise IT strategy, explained that the failures stemmed from contradictory knowledge articles within the company's underlying data rather than a flaw in the AI agent's design.

To resolve the issue, the company repurposed the tool as an auditor agent to detect content anomalies across its public site. After cleaning the problematic data, Salesforce restored the agent to its original functionality.

During a roundtable at Fortune's Brainstorm Tech conference, executives from other firms highlighted the broader implications of the failure. Leaders from Intuit and Amgen argued that poor data foundations and the inability to scale projects from pilots to enterprise-wide adoption are the primary reasons corporate AI initiatives often fail to deliver a return on investment.