News
Nearly Half of Organizations Report Unexpected AI Agent Activity, Study Finds
Nearly half of the security and technology professionals surveyed by Salt Security confirmed or suspected that an AI agent had taken an unintended, unexpected, or unauthorized action in their organization over the past year.
In its "State of Agentic AI and API Security 2H 2026" report, released October 13, the cybersecurity company analyzed responses from more than 380 security and technology professionals alongside production telemetry from hundreds of customer environments.
The research found that 49.8% of respondents confirmed or suspected unexpected or unauthorized AI agent activity.
Among that group, more than half reported unauthorized external API calls or transactions, while more than 40% reported agents modifying or deleting data.
AI Agent Adoption Outpaces Security Controls
According to the report, 86.4% of respondents said their organizations operated at least one AI agent, while 59.5% reported running more than 10. Another 76.8% reported using agents capable of chaining actions across other agents, tools, and APIs. However, only 12.5% said their organizations could consistently trace an agent's complete activity, from its initial instructions through connected tools and APIs to the outcome. And more than 60% said reconstructing an agent's previous-day activity would require hours or days of manual work.
"The data shows a gap that is already producing real operational consequences," said Roey Eliyahu, CEO of Salt Security.
"Organizations are deploying agents and granting them access to enterprise data, APIs, and services before they can trace what those agents are doing or stop them automatically when something goes wrong."
Most Organizations Cannot Automatically Stop Rogue Agents
The report also found that only 17.5% of respondents said their organizations could automatically stop an AI agent that began behaving unexpectedly in a production environment. Most organizations relied on manual intervention or taking the affected application offline.
Salt Security also identified risks involving Model Context Protocol (MCP) servers, which allow AI agents to connect with external tools and services.
According to the research, 32.9% of respondents discovered an unauthorized, shadow, or misconfigured MCP server in the past year, while 54.3% confirmed or suspected such infrastructure.
Separately, Salt Labs reported that MCP-related infrastructure appeared in 21.6% of customer environments in August 2026, up from 16.7% in January.
The report also examined AI-generated code, finding that 83.5% of respondents estimated AI-generated or AI-assisted code contributed to at least some application security findings.
Only 6.8% reported resolving those issues within AI coding assistants when the code was generated.