April 23, 2026
4 min read
Blanca Rivera Campos

Securing AI agents with Europe's first sovereign guardrail platform

Giskard has launched Guards, the first independent, EU-sovereign guardrail platform that allows regulated enterprises to secure their AI applications on-premise. Moving beyond the limitations of generic content filters, the platform is context-aware and built specifically for modern AI agents, securing the full execution chain via a Policy-as-Code framework that includes ready-to-use EU AI Act and OWASP Top 10 LLM compliance packs. To showcase this new standard in enterprise AI security, our technical team is hosting an in-depth live session on May 13, 2026.
Giskard Guards

We are announcing the launch of Giskard Guards, Europe’s first sovereign guardrail platform built specifically for enterprise AI applications. Model guardrails struggle with context-blindness and are not built for AI agents. To address this gap, Giskard Guards provides an on-premise, context-aware security layer that inspects the full agent execution chain, including tool calls, parameter validation, and multi-step reasoning. The platform allows technical teams to enforce custom compliance requirements and pre-built regulatory frameworks, such as the EU AI Act and OWASP Top 10, directly within their own infrastructure.

To celebrate this announcement and present how to protect AI agents in detail, we will be hosting a dedicated live session on May 13, 2026.

🛡️ Giskard Guards is live

Why model guardrails fail enterprise AI deployments?

Context-blindness creates false positives. A banking customer reporting credit card fraud gets blocked as a content violation. That's not security, that's a broken product. Up to 40% of blocked requests are false positives with generic guardrails.

Context-blindness creates false positives

Traditional guardrails are not built for AI agents. They're often designed around toy benchmarks (like blocking forget previous instructions) and fail to account for real-world attacks such as multi-step social engineering, context manipulation, or tool-chain exploitation.

Traditional AI governance can't keep up. Your compliance team fills out risk assessment spreadsheets. Your AI team deploys new agents weekly. By the time a policy is documented, the system has already changed three times. This gap is where incidents happen.

Protect your AI agents with context-aware guardrails

Giskard Guards is built differently. Developed by a French company and EU team, it represents the first independent European platform offering context-aware, sovereign AI security.

  • Agentic by design: Guards sees the full agent execution chain (tool call inspection, parameter validation, and multi-step flow control)
  • Context-specific detection: semantic detectors trained on your business domain, not generic keyword matching.
  • Policy-as-Code: compliance rules expressed in OPA/Rego, versioned in Git, deployable in seconds. Pre-built packs for the EU AI Act and OWASP Top 10 for LLMs included.
  • EU sovereign: Deploy in your infrastructure and keep sensitive traffic under your control.

Customization at scale: Policy-as-Code Guardrails

Giskard Guards converts your regulatory requirements into enforceable policies, versionable, and deployable in seconds.

👉 Protect your agents

🪄 [Live Session] Secure AI Agents with Giskard Guards — May 13, 5 PM CEST

To present Giskard Guards and detail our approach to context-aware AI security, we invite you to our upcoming live session on May 13, 2026, at 5 PM CEST.

David Berenstein will run a live session walking through Giskard Guards: how context-aware detection works in practice, how red teaming findings translate into production guardrails, and how to configure Guards for your specific agent types and threat model.

👉 Register here

Live Session - Giskard Guards

See you soon,

The Giskard Team 🐢

Continuously secure LLM agents, preventing hallucinations and security issues.
Book a Demo

You will also like

Phare LLM benchmark V2: Reasoning models don't guarantee better security

Phare LLM benchmark V2: Reasoning models don't guarantee better security

Phare (Potential Harm Assessment & Risk Evaluation) is an independent, multilingual benchmark designed to evaluate AI models across four critical dimensions, or “modules”: hallucination, bias, harmfulness, and vulnerability to jailbreaking attacks. This second version expands our evaluation to include reasoning models from leading providers, allowing us to assess whether these advanced systems represent a meaningful improvement in AI safety.

View post
OpenClaw security issues include data leakage & prompt injection

OpenClaw security vulnerabilities include data leakage and prompt injection risks

This article explores the critical security failures of the OpenClaw agentic AI, which allowed sensitive data to leak across user sessions and IM channels. It examines how architectural weaknesses in the Control UI and session management created direct paths for prompt injection and unauthorized tool use. Finally, it outlines the essential hardening steps and systematic red-teaming strategies required to transform a vulnerable "fun bot" into a secure enterprise assistant.

View post
Agentic tool extraction: Multi-turn attack that exposes the agent's internal functions

Agentic tool extraction: Multi-turn attack that exposes the agent's internal functions

Agentic Tool Extraction (ATE) is a multi-turn reconnaissance attack to extract complete tool schemas, function names, parameters, types, and return values. ATE exploits conversation context, using seemingly benign questions that bypass standard filters to build a technical blueprint of the agent's capabilities. In this article, we demonstrate how attackers weaponize extracted schemas to craft precise exploits and explain how conversation-level defenses can detect progressive extraction patterns before tool signatures are fully exposed.

View post
Get AI security insights in your inbox