Verno Labs is an offensive security company for AI agents. It points custom offensive LLMs at a customer’s own agents, across text, voice and multimodal interactions, to find vulnerabilities before attackers do.
It is built for teams putting agents into production that need to know how those agents can be manipulated, and to watch for the same attacks once the agents are live. The platform follows a cycle of attack, detect and fix.
At A Glance
| Vendor | Verno Labs (Pallma AI Ltd.) |
|---|---|
| Category | AI model security, AI agent security |
| Identities covered | AI agents and their tool calls |
| Deployment | Managed SaaS or self-hosted on AWS, GCP or Azure |
| Agentic AI | Offensive LLMs, Hydra model, AI security agent |
| Funding | Pre-seed led by Marathon Venture Capital |
| Market classification | AI Model Security; also AI Agent Security |
What Verno Labs Does
An agent that can call tools and act on a user’s behalf can be talked into doing the wrong thing, and those weaknesses often only appear over several turns or across sessions. Verno Labs tests for them with attacks that plan, execute and adapt, both before and during deployment, then keeps watching agents in production. The aim is to find a manipulation path in testing rather than in an incident report. Findings map to an attack taxonomy aligned with OWASP LLM, OWASP Agentic and MITRE ATLAS.
Products
Verno Red
Verno Red runs real attacks against AI agents. Campaigns are multi-turn, adaptive and continuous, and work against text, voice and multimodal agents. Each finding is mapped to OWASP, MITRE and CWE, so security teams can see which class of weakness they are dealing with.
Verno Guard
Once agents are live, Verno Guard monitors every interaction and tool call in production. Adversarial behaviour is raised as a structured incident with severity, context and the attack chain that produced it, and multi-step attacks are reconstructed across sessions.
Verno Fix
Verno Fix turns findings from Red and incidents from Guard into prioritised remediations, generates fixes and tracks each one from discovery through to validated closure.
Hydra And The AI Security Agent
Hydra is Verno Labs’ offensive security model. Rather than replaying a fixed list of prompts, it discovers attacks autonomously and probes agents continuously, reaching weaknesses that only show up as a conversation develops. The AI security agent sits on top of the platform as a natural-language interface, so analysts can ask about vulnerabilities and incidents in plain language instead of searching dashboards.
How It Works
| Component | What it does |
|---|---|
| Deployment | Customers choose managed SaaS or a self-hosted install on AWS, GCP or Azure. |
| Python SDK | Agents are instrumented through a Python SDK. |
| Telemetry | Runtime detections from Verno Guard flow back into testing in Verno Red, so what is seen in production shapes the next round of attacks. |
| Outputs | Findings and incidents are sent to ticketing, messaging and SIEM tools or webhooks. |
| Terminal UI | Alongside the web platform, a terminal UI shows status, policies and tool calls from the command line. |
Identities And Systems Covered
| AI agents | Text, voice and multimodal agents, including their tool calls |
|---|---|
| Agent frameworks | LangChain, LangGraph, custom orchestration and custom agents, through the Python SDK |
| Agent platforms | Microsoft Agent 365, tested for prompt injection, sensitive data exposure, permission abuse and unintended agent actions; Gemini Enterprise |
Integrations
| Area | Integrations |
|---|---|
| Ticketing and incident response | Jira, PagerDuty, Incident.io, Nofire, Resolve |
| Messaging | Slack |
| Security operations | SIEM, webhooks |
Company Background
| Legal entity | Pallma AI Ltd. (engineered in Europe) |
|---|---|
| Founded | 2025 |
| Headquarters | London, United Kingdom |
| Leadership | Pavlos Mitsoulis, co-founder and CEO |
| Funding | A reported $1.6 million pre-seed round with technology leaders from AWS, Meta and Google |
| Research | VernoBench, a benchmark for prompt extraction evaluation |
Note: This overview is compiled from Verno Labs’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.