Vero aims to enforce AI security policies in real time

Vero Security raises $11 million seed to turn AI security policies into active controls during runtime. The startup also focuses on an open community around AI Security Engineering.

Vero Security, a startup founded in 2025, has raised $11 million in seed funding to enforce security policies directly during the execution of AI tools and agents in enterprises. The round, led by TLV Partners with GTM Capital, Maple Capital, and angel investors, also supports the launch of AISecurityEngineering.org, an open community dedicated to this new discipline. The challenge is simple to state but harder to achieve: prevent security rules from remaining just nice documents while agents continue to act.

Quand la politique doit survivre à l’agent

Vero advocates a clear idea: AI security is not limited to taking inventory or producing a remediation report. The company talks about “AI Security Engineering,” meaning the continuous work of discovering how AI is used, assessing risks, then keeping protections up to date. Its software translates decisions approved by security teams into concrete controls, then verifies that corrections have been applied.

The starting point is familiar to many security managers: a validated policy is just a snapshot at time T. A new access, an updated tool, or an added permission can change the risk surface afterward. Vero therefore wants to intervene at the moment the agent attempts an action, not later when potential damage is already visible. It’s less glamorous than a dashboard but much more useful.

Le Runtime Governor, pièce centrale du dispositif

At the heart of the platform is the AI Runtime Governor. Vero explains that it inspects prompts, tool calls, and supported agent actions at the moment the operation is attempted. The goal is to enforce policy during execution, for example by preventing a development agent from reading secrets without reason, blocking an out-of-scope action, or cutting short an attempt triggered by prompt injection.

Prompt injection refers to malicious or manipulative content that pushes an AI system to deviate its behavior. In a simple writing assistant, the effect is often limited to a false or biased response. With an agent that has access to code repositories, credentials, or system commands, the same technique can become a real vector of action on infrastructure. Vero thus emphasizes a key difference: telling a system not to touch a sensitive resource is not the same as technically preventing access.

Pourquoi les contrôles de configuration ne suffisent plus

Vero’s product documentation describes four related functions: inventory, posture analysis, behavioral detection, and enforcement. Inventory identifies agents and their owners. Posture analysis examines configuration and accessible credentials. Behavioral detection observes what happens during the session. Finally, enforcement applies the policy. Together, these components target environments where AI usage moves quickly between endpoint, browser, SaaS, and cloud.

Vero says it integrates with existing security, identity, and endpoint management tools without installing anything new on employee devices. The actual scope depends on integrations and execution paths covered. In an example provided by the company, a Fortune 1000 client defined a rule limiting AI agents’ connection to code repositories and code execution to developers only. Vero claims to have converted this rule into more than 15 controls covering code agents, identities, Model Context Protocol servers, skills, and runtime actions within minutes.

Un pari commercial et un pari de marché

The funding comes with a clear message for CISOs: AI security is no longer just an audit issue but continuous operation. Vero bets on this idea to attract companies deploying agents capable of reading files, calling tools, or executing code. The company was founded by Elad Tessler, CEO, and Daniel Meirov, CTO; the two met within Unit 8200, where they co-led a cyber operations unit of over 100 people, according to the announcement.

The startup already counts several Fortune 500 companies among its users. GTM Capital explains it invested after a structured evaluation with its community of CIOs and CISOs from large enterprises. Vero plans to use the funds to develop the product, expand coverage of enterprise AI technologies, and strengthen its commercial presence in the United States. All with a clear ambition: to ensure security decisions don’t stop at the whiteboard.

Une communauté ouverte pour structurer le métier

Alongside the funding, Vero launches AISecurityEngineering.org, presented as a knowledge center and open community. The platform already offers nine tracks and 259 topics on agent operation, threats, inventory, configuration, identity, runtime defenses, drift monitoring, and incident response. Modules on managing an agentic security program and red teaming are announced as “coming soon.”

The company also funds and operates this site under a public charter stating that content includes no promotional material and member names are not shared with the sales team. The choice is quite clever: at a time when the market is still finding its grammar, providing a working base can help establish vocabulary, methods, and, incidentally, future buying habits.

Key points

  • Vero raises $11 million in seed funding.
  • TLV Partners leads the round, with GTM Capital and Maple Capital.
  • The startup already serves several Fortune 500 companies.
  • The Runtime Governor acts during AI action execution.
  • AISecurityEngineering.org launches nine tracks and 259 topics.
  • The system targets enterprise AI security in 2025 [to be verified].

By the numbers

  • $11 million — amount of the seed round, according to Vero.
  • More than 15 controls — created from a client policy, according to the provided example.
  • 9 tracks — offered at the launch of AISecurityEngineering.org.
  • 259 topics — covered by the public roadmap, according to Vero.
  • More than 100 people — staff of the cyber unit co-led by the founders, according to the announcement.

To read