<meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>HF Demo</title> <style> body { margin: 0; font-family: Arial, sans-serif; } header { background-color: #2c3e50; color: white; text-align: center; padding: 20px; font-size: 28px; font-weight: bold; } </style>

Shadow AI is already inside your company. Here’s how to get control of it

Shadow AI is already inside your company. Here’s how to get control of it

September 7, 2026 - 2:22 pm

Abstract: Close up of Mainboard Electronic computer background. (artificial intelligence, ai)

In a nutshell: Reco’s State of Agent Security 2026 report reveals that 80% of AI tools operate without IT oversight, with SMBs averaging 414 unsanctioned AI tools per 1,000 employees. IBM research indicates that shadow AI adds $670K to breach costs. This article offers guidance on triaging the situation: map access permissions, identify ownership, watch for orphaned agents, and prioritize agents handling customer data, code, and production systems.*

Companies are discovering AI agents integrated into software without IT oversight. But finding them is just the beginning.

For a decade, IT teams could track the full list of software within an organization. Today, AI makes this challenging. A marketing manager might activate an AI feature in existing software; a developer can connect an assistant to an internal knowledge base; or someone could install an AI meeting tool with a simple "Allow" click.

"The first scan often reveals AI hidden in places the organization didn't consider part of its AI program," says Ofer Klein, co-founder and CEO of Reco, a company specializing in enterprise agent security. This includes browser extensions, meeting tools, productivity suites, CRM workflows, support tools, developer environments, and app-to-app integrations.

Reco’s State of Agent Security 2026 report highlights the extent of the issue:

  • Four in five AI tools observed lacked IT oversight.
  • SMBs averaged 414 unsanctioned AI tools per 1,000 employees.

While discovering these tools is crucial, it also raises a new challenge: managing 400 previously unknown AI tools.

The first step should be to assess each agent's actual access:

Instead of immediately shutting down every tool, consider its functionality within the business. Some might perform essential tasks; others might be redundant or risky.

Klein notes that a seemingly harmless assistant could have permissions to:

  • Read emails
  • Summarize files
  • Access customer records
  • Interact with ticketing systems
  • Connect to source code repositories

IBM's 2025 Cost of a Data Breach report, which analyzed 600 breaches across 17 industries, found that one in five involved shadow AI. Organizations with significant shadow AI faced average breach costs $670,000 higher than those with minimal or none.

Leave a Reply