DKDKCISSP
AI SecurityINCIDENT

AI agent API security hasn't caught up with agents already running in production

Postman found that many organizations using AI agents lack adequate security measures for their APIs, raising concerns over unauthorized access.

DKCISSP News Desk20 Nov 2026, 11:00 am3 min read
AI agent API security hasn't caught up with agents already running in production
Image courtesy of Help Net Security. Original report

A recent survey by Postman revealed that a significant number of organizations utilizing AI agents in production have not implemented proper security measures for their APIs. This oversight could lead to unauthorized access to sensitive data and systems.

Postman's survey of nearly 9,000 developers, architects, and executives uncovered that about a quarter of the organizations have AI agents actively calling their APIs in production. However, the survey also highlighted a troubling trend: many companies have not established clear protocols for managing the credentials and identities of these agents, which could lead to significant security vulnerabilities.

The survey indicated that a considerable number of respondents, around four in ten, have not defined how credentials and identity should function for their AI agents. Furthermore, three in ten organizations allow these agents to log in using the same credentials as any other API client. This practice poses risks, as it means that if an agent accesses data or endpoints it should not, there is no distinct identity to limit or revoke its access.

AI agents can be thought of as automated entities capable of interacting with various systems and services within an organization. They can read data, call APIs, and act independently, similar to a new employee who hasn't gone through the necessary onboarding processes. Without proper identification and management, these agents can potentially access sensitive information without authorization.

The survey results showed that many companies prioritize security controls for their APIs over those for the agents themselves. About a third of the respondents reported having no formal process for managing these agents, and fewer than a quarter maintain an inventory of the agents they operate or monitor their activities. Alarmingly, only 9% have a defined process for retiring an agent, meaning most organizations can activate an agent without a plan for its deactivation.

Postman categorized respondents based on their API-first approach, which refers to organizations that design their APIs before developing the surrounding software. Companies that identify as API-first tend to have better organizational practices, such as maintaining an inventory of agents and assigning ownership for each agent. However, even among these companies, a significant percentage, 43.2%, allow their agents to share ordinary credentials.

Despite the growing use of AI agents, only 13.2% of organizations reported that their APIs are adequately prepared for agent interactions. Most APIs have traditionally been designed for human developers, who can interpret documentation and communicate with colleagues for clarification. This poses challenges for AI agents, which rely on accurate and updated documentation to function effectively.

Looking ahead, respondents expressed concerns about the potential for AI-powered attacks and rogue agents misusing APIs, ranking these as their top worries for the next two years. This reflects a growing awareness of the risks associated with deploying AI agents without the necessary security measures in place.

Postman recommends that organizations give each AI agent its own identity to mitigate risks. Currently, about a third of the companies surveyed have already implemented this practice, but many still face significant challenges in managing their AI agents effectively.

What security teams should do

  • Organizations should establish clear protocols for managing AI agent credentials and identities.
  • Security teams need to maintain an inventory of AI agents and monitor their activities.
DKCISSP VIEWOur take, not reported fact

What it meansMany organizations using AI agents lack adequate security controls for their APIs, which could lead to unauthorized access to sensitive data.

Ask your IT team this weekDo we have a process to manage AI agent credentials and identities?How many AI agents are currently operational, and are they monitored?

DecisionThe IT team will implement a protocol for managing AI agent identities, and security will verify compliance.

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 pmPalo Alto Networks to acquire Applied Compute for $4.5 billion19 Nov 2026, 3:53 pmAI agents inherit employee access, and nobody's watching how they use it19 Nov 2026, 10:30 amQuality Isn’t The Main Worry About AI Pentesting — Data Privacy and Production Safety Are18 Nov 2026, 4:33 pm