DKDKCISSP
AI SecurityINCIDENT

AI agents inherit employee access, and nobody's watching how they use it

AI agents built by employees often use their creator's credentials, raising security concerns.

DKCISSP News Desk19 Nov 2026, 10:30 am2 min read
AI agents inherit employee access, and nobody's watching how they use it
Image courtesy of Help Net Security. Original report

AI agents created by employees in various business platforms operate using the access credentials of their creators, according to a report by Nudge Security. This practice poses significant security risks as organizations struggle to monitor and manage these agents effectively.

The report, titled 'AI Adoption in Practice 2026', reveals that within organizations utilizing AI agents, the median number of agents per organization is seven, with about 44% having more than ten. These agents operate across 15 different platforms, complicating security oversight as teams must navigate multiple administrative consoles and permission models.

AI agents often run on the OAuth grants or API keys of the employees who created them, meaning they inherit the same access levels. For example, an agent developed by a sales manager in a CRM can access all data that manager can, which raises concerns about unauthorized access to sensitive information. Since these agents function within the platforms themselves, traditional endpoint security measures may not detect their activity.

OAuth is a protocol that allows applications to access user data without sharing passwords. When employees create AI agents, they typically grant these agents permission to access their emails, files, or other information by clicking 'allow' during setup. Once granted, these permissions remain active until manually revoked, allowing agents continued access even if the creator leaves the company or changes roles.

The report highlights that 74% of organizations had at least one integration with Anthropic's AI, while 67% had OpenAI integrations. Google Workspace was identified as the most common endpoint for these tools. The prevalence of these agents suggests a significant number of organizations may be unaware of the extent of their AI tool usage and the potential risks involved.

Nudge Security found that API keys and access tokens pose the most direct risk, as they can be exploited if they leave the organization's environment. The report recalls a previous incident in 2025, where attackers used stolen OAuth tokens from Drift's AI chat tool to access data from multiple companies' Salesforce environments.

According to Jaime Blasco, CTO of Nudge Security, the increasing integration of AI tools into everyday business platforms means that organizations must be proactive in managing the access these tools utilize. He noted that the gap between what an organization's identity provider tracks and what is actually accessing its data is likely to widen as more platforms introduce agent-building features.

The report concludes with recommendations for organizations to inventory the AI agents present in their systems and assess the access rights associated with each.

What security teams should do

  • Organizations should inventory AI agents built within their platforms and determine whose access each one uses.
  • Review OAuth grants given to AI tools and revoke any that are no longer needed.
DKCISSP VIEWOur take, not reported fact

What it meansAI agents often use the access credentials of their creators, complicating monitoring and increasing security risks.

Ask your IT team this weekWhich AI agents are currently in use across our platforms?What OAuth grants have been issued to AI tools, and are they still necessary?

DecisionThe IT team will inventory AI agents and their associated access rights, and security will verify the findings.

MORE IN AI SECURITY

More cybersecurity reporting

Your AI red team may be finding the same prompt injection thousands of times20 Nov 2026, 12:00 pmAI agent API security hasn't caught up with agents already running in production20 Nov 2026, 11:00 amPalo Alto Networks to acquire Applied Compute for $4.5 billion19 Nov 2026, 3:53 pmQuality Isn’t The Main Worry About AI Pentesting — Data Privacy and Production Safety Are18 Nov 2026, 4:33 pm