Security researchers have warned that major gaps in IT oversight, surging numbers of published vulnerabilities, and MCP security risks are making the AI agent ecosystem increasingly risky.
AI security vendor Reco analyzed anonymized platform telemetry from large enterprises, publicly available Model Context Protocol servers, and vulnerability disclosures from the National Vulnerability Database to compile its report, The State of Agent Security 2026.
It found that 80% of AI tools operate with no oversight, while in SMBs, there are an estimated 414 unsanctioned tools per 1000 employees.
“AI agents have moved from experimentation into daily business workflows, but our findings show only 20% of AI tools in enterprise ecosystems are currently governed by IT oversight,” said Ofer Klein, CEO of Reco.
“That leaves organizations exposed to a new class of operational risk. Agents embedded in applications can operate through existing permissions, OAuth grants and workflow access, creating toxic combinations that expose data and trigger actions beyond what any owner approved.”
Read more on AI agent security risk: AI Agents Now the Enterprises Fastest Growing Exposed Attack Surface
The lack of oversight identified in the report is particularly concerning given Reco’s analysis of 500 MCP servers, which connect AI agents to data and actions.
“Exactly half can execute shell commands directly, which turns a prompt-injection trick into operating system access. More than eight in ten can read or write local files, and roughly three-quarters can make outbound network calls,” the report noted.
“These are the tools agents are built to load, by the thousands, often through a marketplace with no review step. They rarely guard themselves either: just over a quarter expose a network endpoint rather than running locally, and half of those ship no authentication at all, a remotely reachable tool with host-level reach and no lock on the door.”
Worse still, Reco found that nearly two-thirds (62%) of agents combine command execution, file access, and network egress in a single package. This combination can create an end-to-end toolkit to “find data, act on it, and move it off the machine,” Reco claimed.
AI-Related Vulnerabilities Keep Increasing
The report also tracked 637 vulnerabilities across agents and LLM tools.
Of these, 525 were disclosed in the past 18 months, including at least 111 rated critical with CVSS scores of 9.0 or higher. The average monthly disclosure rate rose from less than five during 2023 and 2024 to around 29 since January 2025.
Reco warned that vulnerabilities are being published faster than patching programs can absorb.
