Application of custom rules to automated code review tools to catch issues unique to AI-generated code increased by 10%.

Use of risk-ranking methods to determine where LLM-generated code is safe to deploy increased by 12%.

In 2025, an AI agent placed in the top 5% of teams in a major cybersecurity competition.

65% of enterprises consider action-level guardrails and runtime controls to be a critical priority for AI agents.

Only 21% of enterprises report full visibility into agent actions, MCPs tool invocations, or data access.

30% of businesses reported that they use AI and must now protect it like any other critical system.

AI applications experienced a 43% increase in security incidents over the past 12 months, marking the second-largest increase across all channels.

76% of respondents reported that autonomous AI agents are the hardest systems to secure.

53% of IT leaders globally are highly or extremely concerned about AI security risks.

68% of organizations invested in AI-powered protection capabilities in 2025

74% of organizations expect their focus on AI security to increase significantly over the next two years.

78% of CISOs lack a formal strategy for handling AI identities in a zero trust security architecture in 2025.

100% of organizations plan to invest more of their budget in AI-related security initiatives in the next 12 months.

78% of U.S. CISOs expect AI to create a moderate or significant amount of new IT or security work for their teams due to AI-related security risks and vulnerabilities.

69% of healthcare IT leaders feel pressured to adopt AI faster than they can secure it.

83% of healthcare IT and compliance leaders have raised concerns about AI security.

14% of organizations using Amazon Bedrock do not explicitly block public access to at least one AI training bucket

77% of organizations have the overprivileged default Compute Engine service account configured in Google Vertex AI Notebooks

91% of Amazon SageMaker users have at least one notebook that, if compromised, could grant unauthorized access

Approximately 70% of cloud AI workloads contain at least one unremediated vulnerability