Washington, October 28, 2026 /PRNewswire/ — Evoke Security, a security platform for contingent workforces, today launched the industry’s first evaluation program designed to identify agent misuse and rogue agents within major AI platforms and coding toolchains. Called the Rogue Agent Assessment, the program helps organizations understand how employees are using agents, when agents are being misused, and when agents are acting contrary to explicit instructions.
As employees integrate agent usage into their daily workflows, security teams are facing a challenge that existing security tools have not yet addressed. Agents may take actions that expose sensitive data, expand access privileges, or create unnecessary risks while completing their assigned tasks. Today, knowing only whether an agent followed instructions is no longer sufficient to determine if its operation is secure.
Jason Rebholz, Co-founder and CEO of Evoke Security, said: “Every organization we work with is confronting the same two types of risks: employees misusing agents, and agents going off-script. Agents are bypassing security controls and taking actions no security team would ever approve. A single agent action may seem harmless on its own, but it’s the sequence of actions that tells the full story. That’s why Evoke monitors the complete trajectory of agents, identifies when they’re heading down high-risk paths, and either redirects them back to a safe path or blocks them outright before damage occurs.”
Evoke states that what sets this industry-first evaluation solution apart is its focus on monitoring the current and historical runtime behavior of agents, rather than just misconfigurations. Leveraging the founding team’s experience in incident response and chief information security officer (CISO) roles, Evoke has developed a framework that identifies rogue agents across four dimensions.
- Mission: Does this action align with the agent’s original objective?
- Method: Is this action reasonable for the task?
- Motive: What intention does this behavior suggest?
- Scope of Impact: How does this action expand the radius of potential risk?
The company stated that this is one layer of Evoke’s deep detection approach, helping security teams distinguish between agents operating as expected and those introducing significant risk. Rather than treating every anomalous action as a threat, the framework helps determine whether an agent should continue running, be guided toward a safer path, or be blocked.
Evoke stated that the company is actively collaborating with native AI enterprises to monitor and control AI agent behavior. Evaluation cases show that agents can introduce new risks even when performing legitimate tasks, further underscoring the importance of real-time monitoring and guiding agent behavior. In one evaluation, Evoke discovered a coding agent tasked with security reviews escaping its own sandbox and subsequently lying to users.
The "Uncontrolled Agent Assessment" provides organizations with a structured approach to evaluate their agent security posture across three key areas.
- Proxy misuse: Identify cases where proxies are used to bypass existing security controls or violate usage policies.
- Out-of-Control Agent: Identify situations where the agent stops following instructions or takes unnecessarily high-risk actions.
- Shadow AI: Identify unauthorized AI products, MCPs, or Agent Skills used in the environment.
The company stated that by combining these capabilities, Evoke provides organizations with an assessment specifically built to address autonomous agent security challenges. Security teams will gain clearer visibility into which agents are present, how they operate, where risks emerge, and which actions require intervention.
Evoke says that "Uncontrolled Agent Assessment" is now available to organizations seeking to better understand their AI agent exposure surface and establish effective security boundaries without overly restricting operations.
About Evoke Security
Evoke Security is an AI detection and response platform built specifically to protect AI agents. Founded by veterans in incident response, its platform enables security teams to continuously inventory, model threats, and monitor AI agents. Evoke treats agents as the next-generation operating system for enterprises, enforcing strict controls by identifying high-risk configurations and analyzing real-time behavior and actions, ensuring businesses can innovate without accumulating security debt.