Runtime security startup Eve Security Inc. wants to help stop dangerous artificial intelligence agents in their tracks after raising $4.5 million in funding today.

The round, which was led by Run Ventures and saw participation from Dreamit Ventures, Blu Ventures and LiveOak Ventures, extends the startup’s total seed funding round to $7.5 million.

Eve Security has developed a runtime security layer that provides organizations with the observability, governance and controls they need to deploy AI agents safely and then identify any high-risk or anomalous activities. In the event that it spots an agent misbehaving, it gives teams a way to step in and intervene, before that agent can start meddling with any critical systems.

It’s a totally different security challenge from the old days. Traditional cybersecurity tools were designed for human users and monitoring things like identities, endpoints, applications and infrastructure. Though they can spot things such as a compromised security credential or a suspicious command, they cannot determine if a series of seemingly legitimate actions by an autonomous agent is a normal execution task or laying the groundwork for something malicious.

This is becoming a concern for enterprises, as many are bowing to the inevitable and deploying AI agents even though they still don’t really understand the risks of doing so, simply because they’re afraid of being left behind by their rivals. However, many are getting nervous in the wake of OpenAI Group PBC’s recent disclosure that a number of AI models undergoing a security evaluation were able to escape their testing environment and exploit a previously unknown zero-day vulnerability, allowing them to reach the internet and then compromise Hugging Face Inc.’s production infrastructure.

Eve Security co-founder and Chief Executive Nadav Cornberg said this incident is evidence that agents are becoming extraordinarily capable at a time when they’re being given access to increasingly valuable enterprise systems. “The OpenAI incident showed what can happen when an agent’s pursuit of an objective takes it somewhere its creators never intended,” he explained. “You cannot secure that world simply by deciding in advance what an agent should and shouldn’t do. You have to understand what it is doing, why it is doing it and have the ability to intervene while it is happening.”

The demand for this became obvious during Eve’s fundraising process, when its main backer Run Ventures introduced it to a diverse group of Chief Information Security Officers who validated the need for a new runtime security layer specifically for AI agents. “We had high conviction in the technical thesis, but what we wanted is market proof,” Cornberg said. “We are now seeing that pull directly from CISOs and customers.”

Eve Security launched in January, but since that launch it has had to evolve its platform continuously to keep up with the frenetic pace of agentic innovation. In recent months it has added new capabilities such as discovery, enforcement and automated remediation across Databricks, Microsoft Copilot Studio, Amazon AgentCore, Amazon Bedrock and Glean environments. It has also introduced support for session tainting, enabling it to restrict what agents can do in real-time, based on their prior actions and how exposed they are to sensitive data and systems.

Eve’s platform works by transforming security policies into a deterministic enforcement layer, where policy-matched agent requests and actions are evaluated in real time and enforced deterministically. It also provides even deeper analysis when relevant, enriching its observability data with real-time context from identity providers, data platforms such as Databricks and Snowflake and data loss prevention systems.

Run Ventures partner PT Ungvichian said Eve Security is at the forefront of a new category within the cybersecurity industry. “We believe there will be multiple important companies built in this market,” he stressed. “Eve recognized early that securing autonomous systems requires understanding and controlling behavior at the runtime, and the company is exceptionally well positioned to capitalize on that shift.”

Image: Eve Security