Palo Alto Networks report finds organizations struggle to track and secure AI agents in the cloud
A new report out today from cybersecurity company Palo Alto Networks Inc. finds that artificial intelligence agents have quickly become a fixture in software development, but many organizations can’t properly track or secure what those agents are doing in the cloud.
Now in its sixth year, the State of Cloud Security Report 2026 is based on a survey of 2,800 security executives and practitioners in 10 countries. Along with how AI is being used to build and deploy software, the report looks at how quickly organizations are able to respond when something goes wrong.
The use of AI agents was found to be close to universal, with 99% of respondents saying they use agents for autonomous or semiautonomous actions in their development pipelines. Some 73% said the agents are allowed to modify code, pipeline configurations or production environments, and agents are even doing security testing at 44% of organizations.
The amount of change is going up as a result. Some 82% said AI-driven development has increased the volume of infrastructure updates reaching their cloud, yet organizations on average enforce fewer than three of the eight predeployment security checks measured in the survey. AI is also being used to generate infrastructure-as-code templates at 42% of organizations, but only 34% scan that code before production.
Not surprisingly, problems are getting through. Respondents said that on average, 19% of high and critical application security issues reached production over a 90-day period. It takes an average of 18.4 days to get a fix for a security issue found in code into production.
When respondents were asked what was stopping them from fixing issues before deployment, the most common answer was business pressure to release quickly at 39%. Unclear risk acceptance policies came second at 30%.
Accountability is also an issue when it comes to agents. Two in five organizations said they are unable to reconstruct what an AI agent did, a share that rises to 46% in financial services, while 53% said agent activity is logged under a developer’s identity or personal access token. That means when something goes wrong, investigators first have to figure out which actions were taken by a person and which were taken by an agent before they can work out what was affected.
Removing access once something does go wrong is also slow. Some 60% of organizations still need manual intervention to revoke access for nonhuman identities, and it takes an average of 2.5 days to revoke or rotate compromised developer credentials, API keys or pipeline secrets.
Data governance has its own issues. Nearly all organizations, 99%, said they had found ungoverned AI in their environment in the past year, yet only 9% are able to automatically discover and classify it. Some 40% said governed data had been copied into ungoverned AI tools, and 39% had seen sensitive or regulated data used to train or fine-tune models without authorization.
Despite the gaps, confidence is high. Almost every respondent, 97%, said they are at least somewhat confident they can secure their AI supply chain and applications, though only 36% can monitor AI behavior in production and just 30% can detect prompt injection.
The final section of the report looks at the threat from frontier AI models that can find and exploit software vulnerabilities on their own. Nearly all respondents, 98%, said they are concerned about it, and 65% said it poses a significant or severe threat to global critical infrastructure and financial systems over the next 12 to 18 months. Utilities were the most likely of any sector to rate the threat as significant or severe at 79%, compared with 48% in telecommunications.
Interestingly, though the report refers to “frontier AI-powered capabilities,” the survey questions at the back of it call them “Mythos-grade” threats, a reference to Anthropic PBC’s Claude Mythos models. Palo Alto Networks was one of the launch partners in Anthropic’s Project Glasswing, which gave defenders early access to Claude Mythos Preview in April.
The report argues that against attackers working at machine speed, an 18.4-day fix cycle leaves organizations exposed for more than 17 days after a vulnerability can be discovered. Part of the problem, according to the report, is fragmented tooling, with only 19% of organizations saying they operate with unified telemetry and workflows across code, cloud, identity and runtime. Nearly half of organizations, 47%, still do the work of correlating application vulnerabilities with runtime threats by hand, a number that’s even higher in the U.S. at 55%.
The report ends with a series of recommendations for organizations to secure their cloud environments as AI agents take on more work. Agents should be given their own identities with a clear owner and only the permissions they need to do their job. Security teams are also advised to automate the fixing of recurring issues, though only after rollback procedures have been put in place.
Photo: Palo Alto Networks
A message from John Furrier, co-founder of SiliconANGLE:
Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.
- 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more
- 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network
Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/
About SiliconANGLE Media
Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.