Zenity, a leader in end-to-end security and governance for AI agents, has launched runtime protection for OpenAI’s AgentKit, providing enterprises with real-time enforcement to prevent data leakage, secret exposure, and unsafe agent behavior.

The new capability enables endpoint-level monitoring of every interaction between users and AI agents developed with OpenAI’s AgentKit, instantly detecting and stopping risky or noncompliant actions before they cause harm.

Addressing Security Gaps in AI Agent Development

The announcement builds on findings from Zenity Labs, which recently revealed that OpenAI’s AgentKit could be exploited through prompt injection, response obfuscation, credential exposure, and other sophisticated attack methods. These vulnerabilities could bypass existing guardrails, putting sensitive enterprise data at risk.

Cyber Technology Insights : Inversion6 Launches Growth-Fueled Security Operations Center and MDR Platform

By introducing runtime protection, Zenity strengthens AgentKit deployments with deterministic, policy-driven enforcement—an approach that inspects agent intent and behavior in real time and blocks any unsafe or unauthorized responses before they reach the user.

Understanding OpenAI AgentKit

OpenAI’s AgentKit empowers developers to create and deploy autonomous AI agents using Agent Builder, ChatKit, and the Connector Registry. While these tools accelerate innovation, they also expand the attack surface, leaving potential blind spots where built-in guardrails may fail to recognize complex or nuanced security threats.

As organizations increasingly use AgentKit for both internal and customer-facing workflows, the need for robust AI security and compliance controls becomes critical.

Bridging the Guardrail Gaps with Policy-Based Controls

Zenity’s runtime protection inspects all interactions between users and AgentKit-powered agents, applying rule-based enforcement rather than probabilistic models. This ensures predictable, transparent, and enforceable safeguards across all AI agent communications.

Cyber Technology Insights : eMazzanti: SMBs Urged to Adopt Dark Web Monitoring Amid Rising Cyber Threats

Key features of Zenity’s runtime protection for OpenAI AgentKit include:

  • Data Leakage Detection – Identifies and prevents any attempt by AI agents to exfiltrate confidential or regulated information.
  • Secrets Exposure Prevention – Detects embedded API keys, credentials, or tokens within responses and blocks exposure in real time.
  • Unsafe Response Blocking – Stops responses that violate company policies, compliance requirements, or trust standards before delivery.

A Secure Foundation for the Future of AI Agents

“AgentKit revolutionizes how AI agents are created and scaled, but it also dramatically widens the potential attack surface,” said Michael Bargury, CTO and Co-Founder of Zenity. “Our research has shown that existing guardrails can overlook critical risks—from subtle prompt injections to hidden data leaks. Zenity’s runtime protection closes that gap by analyzing every interaction, understanding intent, and enforcing strict security policies before harm can occur.”

With this new capability, Zenity positions itself as a vital partner for enterprises seeking to secure AI agent ecosystems without slowing innovation. The platform’s deterministic enforcement model ensures consistent protection while maintaining developer agility across AI-driven environments.

Cyber Technology Insights : Surfshark Protects Against Phishing Attacks with The Email Scam Checker

To participate in our interviews, please write to our CyberTech Media Room at info@intentamplify.com