Zenity, a leader in end-to-end security and governance for AI agents, has just released runtime protection for OpenAI AgentKit. This is to support companies in securing agentic AI workflows.
The newly added feature in Zenity’s product line closes security gaps in AgentKit’s built-in guardrails by monitoring agent behavior on endpoints. It also implements rigid, policy-based rules. It detects and blocks data leakage, secret exposure, and unsafe agent responses in real time.
According to Zenity’s research team, AgentKit’s safety layers were shown to be bypassable via techniques such as prompt injection, response obfuscation and credential leakage.
Zenity’s offering works at the user-agent interaction layer. It monitors each response before delivery, evaluates intent, and applies deterministic controls rather than probabilistic guardrails. The key protections include
- Data leakage detection: blocking attempts by agents to exfiltrate sensitive or regulated data.
- Secrets exposure prevention: identifying embedded credentials and preventing their disclosure.
- Unsafe response blocking: stopping delivered content that violates compliance, policy or brand trust.
Michael Bargury, CTO and co-founder of Zenity, noted that while AgentKit speeds agent creation and deployment, it also massively expands the attack surface. He said Zenity’s runtime protection fills that crucial gap.
The use of agentic AI by enterprises has risen in both internal and customer-facing workflows. So, the need for solid security and governance has become paramount. Zenity’s platform allows companies to experiment with AI agents while simultaneously
putting in place strict security measures and preserving trust.
Overall, this release positions Zenity as a key partner to companies that use AI agents through AgentKit in their safe scaling efforts. This is where they achieve a balance between freedom and control.
To explore how Security Operations Centers (SOC) play a crucial role in defending against modern cyber threats, read our latest SOC News.
Source: Businesswire
