Security

AI Agents Outpace Security Controls in Cloud Environments

Organizations grant code modification rights to 73% of AI agents but enforce fewer than three predeployment security checks on average.

Omega Editorial · November 16, 2026 · 3 min read

Key takeaways

  • 99% of organizations use AI agents in development pipelines, with 73% allowing agents to modify code or production environments directly.
  • Organizations enforce fewer than three of eight standard predeployment security checks despite 82% reporting increased infrastructure updates from AI-driven development.
  • Two in five organizations cannot reconstruct AI agent actions after incidents, and 53% log agent activity under developer identities.
  • High and critical security issues reach production at a 19% rate over 90 days, with an average 18.4-day fix cycle.
  • Only 19% of organizations have unified security telemetry across code, cloud, identity, and runtime environments.

Visibility gaps widen as autonomous agents proliferate

Artificial intelligence agents have become nearly universal in software development environments, but security controls have not kept pace with their adoption, according to Palo Alto Networks' sixth annual State of Cloud Security Report released today.

The survey of 2,800 security executives and practitioners across 10 countries reveals that 99% of organizations now deploy AI agents for autonomous or semiautonomous actions in their development pipelines. Among these organizations, 73% permit agents to modify code, pipeline configurations, or production environments directly. Another 44% have agents conducting security testing.

Yet organizations enforce fewer than three of the eight standard predeployment security checks measured in the study. The disconnect grows more pronounced with infrastructure-as-code: while 42% of organizations use AI to generate these templates, only 34% scan the generated code before it reaches production.

Why it matters

The 18.4-day average time to fix security issues in production creates a dangerous exposure window as frontier AI models gain the ability to discover and exploit vulnerabilities autonomously. With 65% of respondents viewing AI-powered vulnerability exploitation as a significant or severe threat to critical infrastructure within 12 to 18 months, the gap between agent capabilities and security controls represents a systemic risk that compounds as adoption accelerates.

Attribution and accountability challenges

Two in five organizations cannot reconstruct what an AI agent did after an incident, a figure that climbs to 46% in financial services. The problem stems from logging practices: 53% of organizations record agent activity under a developer's personal identity or access token, forcing investigators to separate human actions from agent actions before determining impact.

Access revocation remains slow and manual. Sixty percent of organizations require human intervention to revoke access for nonhuman identities, and the average time to revoke or rotate compromised developer credentials, API keys, or pipeline secrets stretches to 2.5 days.

Security issues reach production at scale

Respondents reported that 19% of high and critical application security issues reached production environments over a 90-day period. The volume of infrastructure updates has increased at 82% of organizations due to AI-driven development.

When asked what prevents fixing issues before deployment, 39% cited business pressure to release quickly, while 30% pointed to unclear risk acceptance policies.

Data governance failures persist

Nearly all organizations (99%) discovered ungoverned AI in their environments over the past year, yet only 9% can automatically discover and classify it. Forty percent saw governed data copied into ungoverned AI tools, and 39% found sensitive or regulated data used to train or fine-tune models without authorization.

Confidence exceeds capability

Despite the documented gaps, 97% of respondents expressed at least some confidence in their ability to secure AI supply chains and applications. However, only 36% can monitor AI behavior in production, and just 30% can detect prompt injection attacks.

The report notes that 98% of organizations are concerned about frontier AI models that can autonomously find and exploit vulnerabilities. Utilities rated this threat most seriously, with 79% calling it significant or severe compared with 48% in telecommunications.

Only 19% of organizations operate with unified telemetry and workflows across code, cloud, identity, and runtime. Nearly half (47%) still correlate application vulnerabilities with runtime threats manually, a figure that rises to 55% in the United States.

Palo Alto Networks recommends giving agents dedicated identities with clear ownership and minimum necessary permissions, automating fixes for recurring issues, and establishing rollback procedures before implementing automated remediation.

These findings were first reported by SiliconANGLE based on Palo Alto Networks' State of Cloud Security Report 2026.

  • #ai agents
  • #cloud security
  • #devsecops
  • #infrastructure as code
  • #nonhuman identities
  • #vulnerability management

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