AI agents are moving beyond chatbots and into everyday business systems — reading emails, creating tickets, writing code and moving data between applications. But their growing access is creating a new and often invisible cybersecurity risk.
A recent report highlighted that four out of five AI tools operate without IT oversight, while smaller companies may have hundreds of unsanctioned AI tools per 1,000 employees. These agents can retain access through OAuth permissions and continue operating without traditional human supervision.
The risk becomes even greater with agentic systems connected to tools and infrastructure. Researchers found that half of 500 analysed MCP servers could execute shell commands, potentially giving compromised or manipulated agents significant access to corporate environments.
The report also identified 525 vulnerabilities in agent and LLM tooling over the previous 18 months, including at least 111 rated critical.
For enterprises, the message is clear: cybersecurity must evolve from controlling what AI generates to controlling what AI is allowed to do.
