Verizon: shadow AI now third most common data leak

Verizon's Data Breach Investigation Report provides the quantitative backdrop for the guidance: nearly a third of all safety breaches begin with vulnerability exploitation, and shadow AI has become the third most common non-malicious data leakage-related activity. The author, who leads Data, AI and Quantum Safe Security Services at a global technology company, frames AI as just another technology from a security context, but notes it is unique in that it can operate autonomously without automatically understanding what is confidential or where its responsibilities should end.

The four practices are built around that gap. The first two focus on employees: use only organization-approved AI tools, and verify AI outputs rather than trusting generated links, documents, or code. The latter two extend to autonomous agents. Agentic AI can carry out entire tasks with limited human intervention, and an example is an invoice-approving agent that, without appropriate guardrails, might keep trying different ways to complete its assigned task even when approval isn't authorized. Employees who oversee agents are asked to understand what tasks and decisions they are authorized to perform, when human approval is required, and how to report or stop unexpected behavior.

Most organizations have incident response policies for traditional cybersecurity threats, but those procedures may not yet fully address AI-related incidents. It encourages employees to familiarize themselves with their company's AI policies and reporting procedures, and to report accidental sensitive-data sharing or unexpected agent behavior promptly to IT or security teams rather than investigating it themselves. It also points to Verizon AI Skills for America, which offers free access to AI training, certificates, and resources designed to help workers prepare for the evolving AI economy.