Giskard
Giskard is a Paris-based AI security company that tests large language models and AI agents for vulnerabilities before they reach production.
Profile
Giskard tests AI agents and large language models for security vulnerabilities and quality failures, then provides a deployment recommendation and remediation guidance.
Giskard is a Paris-based AI security company that tests large language models and AI agents for vulnerabilities before they reach production. Founded in 2021 by Alex Combessie and Jean-Baptiste Cordonnier, the company emerged from research at the French AI startup ecosystem and open-sourced its testing framework on GitHub, where it now has over 5,400 stars. Giskard's core offering is a managed assessment service that probes AI agents for security failures—prompt injection, data disclosure, sycophancy, inappropriate content—and quality issues such as hallucinations, contradictions, omissions, and inappropriate denials.
The company guarantees a refund if no vulnerabilities are found. After each assessment, clients receive a structured report with a go/no-go deployment recommendation signed by Giskard's security experts, a ranked vulnerability list, and remediation guidance that includes opening tickets in the client's workflow. Giskard emphasizes data sovereignty: it offers data residency in the EU or US, role-based access control, audit trails, SOC 2 Type II and HIPAA compliance, and a stated 0-training policy on client data.
The company maintains a public database of real AI incidents called RealHarm. As of mid-2026, Giskard has not disclosed its own revenue or funding rounds; the research dossier contains no financial data, headcount, or investor information for the company. Its customer base appears to include enterprises such as Michelin, based on testimonials on its website, though no other named clients appear in the dossier.
Giskard operates in a rapidly growing market for AI red-teaming and security, competing with startups like Robust Intelligence and larger platform vendors that embed similar testing capabilities. The company's open-source strategy has built developer awareness, but its transition to a paid managed service model will determine its commercial trajectory.
Who buys this
- Enterprises deploying customer-facing AI chatbots or virtual agents
- Financial services firms using LLMs for document processing or advisory
- Healthcare organizations running AI-powered clinical decision support
- Insurance companies automating claims handling with generative AI
- Any regulated industry needing audit trails for AI safety before production release
Strengths and what to watch
Strengths
- Open-source testing framework with 5,400+ GitHub stars provides broad developer awareness and community validation
- Guaranteed refund if no vulnerabilities found, signaling confidence in detection depth and differentiating from consultancies
- Data sovereignty features (EU/US residency, SOC 2, HIPAA, RBAC) address enterprise compliance requirements directly
Watch for
- No disclosed revenue, funding, or headcount makes it impossible to assess financial health or growth trajectory
- Relies on a managed-service model that may struggle to scale compared to self-serve or platform-embedded testing tools
- Customer concentration risk: only Michelin is publicly named; lack of diverse named logos suggests early commercial stage
Key Information
- Industry
- AI Safety & Security
- Founded
- 2021
- Headquarters
- Paris, France
Frequently Asked Questions
What is Giskard and what does it do?
Giskard is a Paris-based AI security company that tests large language models and AI agents for security vulnerabilities and quality failures before production. It provides a deployment recommendation and remediation guidance after each assessment.
How does Giskard test AI agents for vulnerabilities?
Giskard probes AI agents for security failures like prompt injection, data disclosure, sycophancy, and inappropriate content. It also checks quality issues such as hallucinations, contradictions, omissions, and inappropriate denials, then delivers a structured report with remediation guidance.
Does Giskard offer a refund if no vulnerabilities are found?
Yes, Giskard guarantees a refund if no vulnerabilities are found during an assessment. This policy signals confidence in its detection depth and differentiates it from consultancies that charge regardless of findings.
What compliance and data sovereignty features does Giskard provide?
Giskard offers data residency in the EU or US, role-based access control, audit trails, SOC 2 Type II and HIPAA compliance. It also states a 0-training policy on client data, addressing enterprise regulatory requirements for AI safety.
Is Giskard open source and how popular is it?
Giskard open-sourced its testing framework on GitHub, where it has over 5,400 stars. This open-source strategy has built developer awareness and community validation, though its commercial model is a paid managed assessment service.
What industries does Giskard serve?
Giskard serves enterprises deploying customer-facing AI chatbots, financial services using LLMs, healthcare organizations with clinical decision support, insurance companies automating claims, and any regulated industry needing audit trails for AI safety before production release.
Sources
- www.giskard.ai — Product description, vulnerability categories, GitHub star count, data sovereignty claims, refund guarantee, RealHarm database, Michelin testimonial
- ir.crowdstrike.com — CrowdStrike FY26 earnings; no Giskard-specific data, but included in dossier as a potential partner or competitor reference
- www.gsk.com — GSK quarterly results; no Giskard-specific data, included in dossier erroneously
- www.sec.gov — SunCar Technology FY2025 earnings; no Giskard-specific data, included in dossier erroneously