Most enterprises are rolling out frontier AI faster than they are securing it, according to new threat research from Zscaler.

In its ThreatLabz 2026 Report on Frontier AI and Enterprise Readiness, published August 7, 2026, the vendor’s research team scored hundreds of enterprises on a 0 to 100 scale across data protection, attack-surface defense, and decoy deployment. The average came in at 37. The governance pillar scored 2.1 out of a possible 20, which Zscaler Chief Security Officer Deepen Desai calls “year zero.”

The report ties those numbers to specific gaps: ungoverned internal AI endpoints holding proprietary data, identity and access sprawl reaching thousands of applications, and VPN trust chains that let an intruder move laterally in minutes. Zscaler illustrates the risk with a hypothetical breach scenario.

For security leaders, the value is the benchmark. A 37 average and a near-zero governance score give CISOs a concrete figure to bring to boards that are demanding rapid AI adoption without funding matching controls. The findings also name clear audit starting points: inventory and govern internal AI endpoints, measure identity sprawl across your app estate, and stress-test VPN and trust assumptions.

Source: Zscaler ThreatLabz