Quick Buzz Feed
Reco report reveals shadow AI and surge in vulnerability disclosures
Security researchers have warned about significant gaps in IT oversight for AI tools, alongside an increasing number of published vulnerabilities and Model Context Protocol (MCP) security risks, making the AI agent ecosystem more hazardous. A report by AI security vendor Reco, 'The State of Agent Security 2026', found that 80% of AI tools operate without IT oversight in large enterprises, and SMBs show an estimated 414 unsanctioned tools per 1000 employees. This leaves organizations vulnerable to new operational risks, as AI agents embedded in applications can leverage existing permissions to expose data and execute unauthorized actions. Reco's analysis of 500 MCP servers revealed that half can execute shell commands, over 80% can read/write local files, and roughly three-quarters can make outbound network calls, often without authentication. Worryingly, 62% of agents combine command execution, file access, and network egress, creating a potent toolkit for data exfiltration and manipulation.
The report also tracked 637 vulnerabilities specifically related to AI agents and Large Language Model (LLM) tools. A significant portion, 525 of these vulnerabilities, were disclosed within the last 18 months, with at least 111 classified as critical (CVSS scores of 9.0 or higher). The average monthly disclosure rate for these vulnerabilities has dramatically increased from less than five per month during 2023 and 2024 to approximately 29 per month since January 2025. Reco highlighted that these vulnerabilities are being discovered and published at a faster rate than organizations can implement effective patching programs, further exacerbating the security challenges in the rapidly evolving AI landscape.