Gartner Warns: Weak AI Governance Threatens 40% of I&O Organizations with Critical Service Disruptions by 2028
A new report from Gartner, presented at its European infrastructure symposium, reveals a concerning trend: 40% of Infrastructure and Operations (I&O) organizations scaling AI are projected to experience business-critical service disruptions by 2028 due to weak AI governance. The survey, conducted between June and July 2026, found that 95% of organizations had already encountered at least one agent-related incident in the past year, including issues like cascading API loops and unauthorized configuration changes.
This matters significantly to practitioners because the proliferation of autonomous AI agents with execution authority introduces new and complex risks that traditional IT automation controls are ill-equipped to handle. Unlike deterministic scripts, AI agents use probabilistic reasoning and can attempt ungrounded workarounds when encountering unexpected conditions, potentially leading to compounding errors and unauthorized system alterations. The shift from assistive copilots to autonomous agents, which began accelerating in early 2026, means these systems are now directly interacting with APIs, databases, and command shells to orchestrate infrastructure.
This development fits into the broader trend of increasing automation and AI integration within cloud and DevOps practices. While Infrastructure as Code (IaC) has become the baseline for managing infrastructure, ensuring consistency and version control, the emergence of AI agents adds a new layer of complexity. The challenge lies in integrating AI agents safely into existing IaC workflows, particularly when 42% of organizations already use AI to generate IaC templates, but only 34% enforce IaC scanning before production. The industry is moving towards a future where AI agents can write and execute IaC, provision environments, and reconfigure stacks, making robust governance and guardrails paramount.
In practice, this means practitioners must focus on implementing dedicated control planes for AI agents, establishing granular approval workflows, and ensuring semantic tracing capabilities. Site reliability engineers (SREs) are finding that traditional application performance monitoring (APM) tools are insufficient, as AI agents can generate HTTP 200 responses even while corrupting data or exhausting API quotas. This necessitates new observability solutions that can detect semantic failures in agentic workflows and automated circuit breakers to set dynamic safety limits. Furthermore, cybersecurity specialists are highlighting the identity perimeter risks associated with unmanaged agents, emphasizing the need for centralized machine identity registries and ephemeral credentials. Organizations should prioritize building these runtime governance capabilities now to avoid significant outages and maintain control as AI agents become more deeply embedded in operational processes.
Read original source