ThreatDown found that 74% of organizations are exposed to shadow AI, highlighting a widening governance gap as employees adopt unsanctioned AI tools beyond the visibility of IT and security teams.
Shadow AI exposure reveals enterprise visibility gaps
Shadow AI refers to the unsanctioned use of AI tools, models, and browser features by employees without organizational approval or visibility from IT and SecOps teams.
This exposure is caused by organizations operating far more AI tools than they expected.
“That matters because of what it could hide,” ThreatDown states. “Someone on your team pasted a customer contract into a chatbot last week to save a few minutes. Someone else dropped in a chunk of source code to debug it faster. Neither of them thought twice about it. Both of them moved company data onto infrastructure your security team has never seen, never audited, and can’t control – once that data gets out, there’s no pulling it back.”
Different departments within an organization have their favorite tools to use, and they make separate decisions about what enters and leaves an enterprise environment.
Tooling takes action as agents enter the workflow
ThreatDown notes that tools don’t just take input anymore; they act, with many tools running as agents now. These agents act as software with standing permissions to read files, write and run code, and make decisions.
These agents also reach out to other systems through MCP, which creates a new stealth supply chain that lacks oversight.
When a threat actor hijacks a shadow agent, they can gain access to everything that agent was trusted to touch, from files to credentials.
Organizations are running more AI tools than expected
Among the companies ThreatDown surveyed, 60% expected their tool count to be between one and five tools, but in reality, only 26% of companies surveyed were running between one and five tools.
A majority of the companies surveyed are operating between six and 15 tools (44%), and 30% are operating over 16 AI tools.
“If your organization expected a few tools, but is actually running 16 or more, that means most of the AI tools your employees rely on are operating with no oversight as they process company data, execute unreviewed code, or reach out to systems nobody signed off on,” ThreatDown notes.
Employee AI adoption outpaces company estimates
Total tool count isn’t the only area that is contributing to shadow AI exposure.
The surveyed companies estimated that approximately 33% of their workforce was utilizing AI tools, when the actual number was 58%.
This lack of insight into how many employees are using AI is a significant contributor to shadow AI exposure, leaving organizations at increasingly greater risk.
READ MORE: TrustedTech research shows shadow AI usage in its customer base is largely due to leaders not following corporate policy or otherwise disclosing tools.
ThreatDown positions AIDR as a governance tool
To help mitigate the challenges that shadow AI presents, ThreatDown developed AI Detection & Response (AIDR).
ThreatDown AIDR gives IT and security teams visibility into how AI is being used across the entire organization, identifying governance gaps and unauthorized AI use that put the company at risk.
Among the capabilities are:
- AI tool inventory visibility: Visibility into every AI application used within an organization, including shadow AI tools installed without IT approval. This capability aggregates app name, category, platform, vendor, version, and endpoint count in a single view.
- Endpoint AI activity monitoring: View which users are accessing which AI tools, from which devices, and how often. Web protection captures AI activity at the endpoint level to give security teams behavioral context network logs alone don’t provide.
- Unified dashboard, activity feed, and tool overview: Provides AI usage trends, real-time activity events, and full tool breakdowns in a single consolidated view with no context-switching between consoles and no manual correlation across disconnected tools.
- Monitoring built for governance: Newly discovered AI tools are automatically surfaced for review and can be labeled as authorized or unauthorized – and new tools can be set to unauthorized by default.