Palo Alto Networks survey of 2,800 respondents finds 99% of organisations use AI agents in pipelines, yet only 9% can automatically find ungoverned AI
AI agents now sit inside almost every software pipeline, yet most security teams struggle to see what they are doing. That is the central finding of the State of Cloud Security Report 2026 from Palo Alto Networks.
The scale of this problem is massive, as the report reveals that 99% of organisations have found ungoverned AI in their environments in the past twelve months.
The sixth annual study draws on responses from 2,800 executives and practitioners across 10 countries and four regions.
AI agent security risks in the development pipeline
Agents have moved well beyond coding assistance. The report finds 99% of organisations use them for autonomous or semiautonomous actions in development pipelines.
About 73% let agents modify code, pipeline configuration or production, and 82% say AI-driven development has increased the number of infrastructure updates reaching the cloud.
Controls have not kept pace. Organisations enforce fewer than three of the eight predeployment checks measured, while 56% rely on human approval to stop a risky agent action.
Infrastructure as code (IaC) adds pressure. While 42% of organisations use AI to generate IaC templates, only 34% enforce IaC scanning before production, which lands it among the least-enforced security checks in the stack.
Fixes are slow too. It takes 18.4 days on average to deploy a security fix, and 19% of high and critical application security issues reach production within 90 days.
The report puts the challenge plainly: “Cloud-native security now has to account for software that writes software, agents operating through privileged workflows, identities that include autonomous systems and sensitive data moving through AI infrastructure outside established visibility.”
As Wendi Whitmore, the CISO at Palo Alto Networks, writes on LinkedIn: “The future of AI trust won’t be determined by how much we trust the models. It will be determined by the cybersecurity controls we build around them.”
AI agent accountability and credential risks
Accountability is another gap, with 40% of organisations unable to reconstruct an AI agent’s actions.
Access is also part of the problem, with 39% reporting that AI coding tools or agents can reach production cloud environments through development or deployment workflows.
Identity is another. According to the report, 53% of engineers log agent actions under their own identities or personal access tokens, which makes human and machine activity hard to separate.
At a time when threat actors act at machine speed, the report has concerningly found that revoking access happens too slow.
The future of AI trust won’t be determined by how much we trust the models. It will be determined by the cybersecurity controls we build around them.Wendi Whitmore, the CISO at Palo Alto Networks
Some 60% of organisations need manual intervention to revoke nonhuman access, and compromised developer credentials take 2.5 days on average to revoke or rotate.
Visibility is thin as well. While 99% of organisations found ungoverned AI in their environment over the past 12 months, only 9% can automatically discover and classify it.
About 40% have seen governed data copied into ungoverned AI environments and 70% cannot map sensitive data flows across AI systems.
Critical infrastructure threats from frontier AI
Concern is near universal, with 98% of organisations worried about the speed at which AI models can discover and weaponise software vulnerabilities.
According to the report, 65% see a significant or severe threat to global critical infrastructure and financial systems over the next 12 to 18 months.
The report has also found that confidence in AI runs ahead of capability. 97% of respondents are at least somewhat confident in their AI supply chain security strategy, yet only 36% can monitor AI behaviour in production.
The concern over risks from autonomous AI is dependent on sector. Utilities rate the threat highest at 79%, followed by transportation at 72% and business services at 50%.
The report describes the resulting shift: “The security boundary shifts from protecting IT networks to safeguarding critical infrastructure and public safety.”
Its recommendations include giving agents distinct identities, accountable owners and task-scoped permissions. It also advises combining short-lived credentials with rehearsed, granular revocation.
Key figures from the Palo Alto Networks' State of Cloud Security Report 2026
99% of organisations have identified ungoverned AI in their environments over the past 12 months, yet only 9% can automatically discover and classify it
99% of organisations use AI agents for autonomous or semiautonomous actions in development pipelines, with 73% allowing agents to modify code, pipeline configuration or production
56% rely on human approval to prevent risky AI agent actions, while organisations enforce fewer than three of the eight measured predeployment security checks
53% of engineers log AI agent activity under their own identities or personal access tokens, making it difficult to distinguish human and machine activity
60% of organisations require manual intervention to revoke nonhuman access, while compromised developer credentials take an average of 2.5 days to revoke or rotate
- 70% of organisations cannot map sensitive data flows across AI systems, while 65% see a significant or severe threat to global critical infrastructure and financial systems over the next 12 to 18 months
Key partners of Palo Alto Networks
Google Cloud: Works with Palo Alto Networks to deliver AI-powered cloud security, threat detection and protection for enterprise cloud infrastructure.
Amazon Web Services (AWS): Partners with Palo Alto Networks to provide integrated cloud security solutions and help organisations protect workloads across AWS environments.
Microsoft: Collaborates with Palo Alto Networks to strengthen enterprise security through cloud, network and threat intelligence integrations.
Accenture: Combines its cybersecurity consulting and managed services expertise with Palo Alto Networks' security platforms to support enterprise digital transformation.
Deloitte: Works with Palo Alto Networks to deliver integrated cybersecurity services, helping organisations modernise security operations and strengthen cloud protection.
NTT DATA: Has a multi-year strategic alliance with Palo Alto Networks focused on AI security, identity protection, cloud resilience and managed security services.
Wipro: Partners with Palo Alto Networks to deliver AI-driven managed detection and response services through integrated security operations.
IBM: Collaborates with Palo Alto Networks across enterprise cybersecurity, cloud protection and security transformation initiatives.