Security

99% of Firms Found Ungoverned AI, Only 9% Can Track It

Palo Alto Networks survey reveals massive visibility gap as AI agents proliferate across development pipelines without adequate security controls.

Omega Editorial · November 16, 2026 · 3 min read

Key takeaways

  • 99% of organizations found ungoverned AI in their environments over the past year, but only 9% can automatically discover and classify these systems
  • 73% of organizations allow AI agents to modify code, pipeline configuration, or production systems, yet most enforce fewer than three pre-deployment security checks
  • 53% of engineers log AI agent activity under their own credentials, making it impossible to distinguish human from machine actions in security audits
  • Compromised developer credentials take an average of 2.5 days to revoke, while 60% of organizations require manual intervention to revoke non-human access
  • 65% of respondents see significant or severe threats to critical infrastructure and financial systems from AI over the next 12 to 18 months

Nearly every organization has ungoverned AI running somewhere in its infrastructure, yet almost none can automatically find it, according to new research from Palo Alto Networks.

The company's State of Cloud Security Report 2026 surveyed 2,800 executives and practitioners across 10 countries and found that 99% of organizations discovered ungoverned AI in their environments over the past year. Only 9% can automatically discover and classify these systems.

The findings, first reported by Palo Alto Networks in its sixth annual cloud security study, point to a fundamental mismatch between AI adoption velocity and security capability.

Agents operate with broad access and weak oversight

AI agents have moved beyond simple coding assistance into autonomous roles across development pipelines. The research shows 99% of organizations now use agents for autonomous or semi-autonomous actions in development, with 73% allowing agents to modify code, pipeline configuration, or production systems directly.

Yet security controls remain sparse. Organizations enforce fewer than three of the eight pre-deployment security checks measured in the study. More than half—56%—rely on human approval as the primary mechanism to prevent risky agent actions.

Infrastructure-as-code presents particular challenges. While 42% of organizations use AI to generate IaC templates, only 34% enforce IaC scanning before production deployment. Security fixes take an average of 18.4 days to deploy, and 19% of high and critical application security issues reach production within 90 days.

Identity and access controls lag behind AI deployment

Accountability gaps compound the visibility problem. Forty percent of organizations cannot reconstruct what an AI agent did after the fact. More than half of engineers—53%—log agent actions under their own identities or personal access tokens, making it impossible to distinguish human activity from machine activity in audit logs.

Access revocation happens too slowly for the threat environment. Sixty percent of organizations require manual intervention to revoke non-human access, and compromised developer credentials take an average of 2.5 days to revoke or rotate. Meanwhile, 39% reported that AI coding tools or agents can reach production cloud environments through development or deployment workflows.

Data governance shows similar weaknesses. Seventy percent of organizations cannot map sensitive data flows across AI systems, and 40% have observed governed data copied into ungoverned AI environments.

Why it matters

The research reveals a dangerous asymmetry: AI agents operate at machine speed with privileged access across development and production systems, while security teams lack basic visibility and control mechanisms. With 65% of respondents seeing significant or severe threats to critical infrastructure and financial systems over the next 12 to 18 months, the window to establish governance frameworks is closing. Organizations that cannot automatically discover and classify AI systems in their environments are operating with fundamental blind spots in their security posture.

Infrastructure sectors see highest risk

Concern about AI-driven threats varies by sector. Utilities rated the threat highest at 79%, followed by transportation at 72% and business services at 50%. Overall, 98% of organizations expressed concern about the speed at which AI models can discover and weaponize software vulnerabilities.

Palo Alto Networks recommends giving agents distinct identities, accountable owners, and task-scoped permissions, combined with short-lived credentials and granular revocation capabilities.

The findings were published in Palo Alto Networks' State of Cloud Security Report 2026.

  • #ai security
  • #cloud security
  • #ai agents
  • #palo alto networks
  • #identity management
  • #infrastructure security

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Security