AI Invents Security Data: Ivanti Catches Fabrications

AI systems are reportedly inventing data within cybersecurity guidance, a concerning trend that Ivanti has proactively identified. Chris Goettl, Ivanti's VP of product management for endpoint security, trained an AI skill that began fabricating details. This skill, designed to apply Ivanti's threat-risk prioritization to raw patch data, draws from published advisories and internal spreadsheets.

Goettl discovered these fabrications during training, leading Ivanti to implement a mandatory human review step for all AI-generated outputs. This ensures that customers only receive approved and accurate information. The AI skill's output, including prioritized Patch Tuesday briefings, now undergoes human vetting before release.

Recent Patch Tuesday events reveal significant discrepancies in vulnerability counts. For instance, Microsoft's September 8th release, the largest in its history, showed a gap of 205 CVEs between different tracking reports. This highlights the complexity of data interpretation and the potential for AI to err.

Ivanti's approach contrasts with other vendors, as it publicly disclosed the AI fabrications and its human-in-the-loop process. This transparency aims to build trust, particularly as organizations increasingly rely on AI for critical security information. The company's efforts underscore the necessity of human oversight in AI-assisted workflows.