Giskard
Giskard is a French AI security company that provides automated red teaming, LLM evaluation, and runtime guardrails for enterprises deploying conversational AI agents, serving regulated European customers in banking, insurance, retail, healthcare and the public sector via an open-source library and an enterprise Hub.
- Company typePrivate
- Founded2021
- HeadquartersParis, France
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Giskard does
Giskard (operating as GISKARD AI SAS) is a French AI security company founded in 2021 and headquartered in Paris that builds an AI red-teaming and LLM security platform for organizations deploying conversational AI agents. Its product architecture combines a free, open-source Python testing library distributed via GitHub and PyPI (with 1,000+ stars) with a commercial enterprise platform called Giskard Hub, which bundles three modules — Continuous Red Teaming, LLM Evaluation, and Giskard Guards (context-aware runtime guardrails) — and is complemented by a managed AI security assessment service (Giskard Sofia) delivered by Giskard's ML researchers.
The underlying technology is a black-box testing architecture that treats any conversational AI agent as an API endpoint. Giskard uses dynamic multi-turn adversarial agents (notably GOAT, which applies Chain-of-Attack-Thought reasoning), LLM-as-a-Judge meta-evaluation, and a runtime guardrail layer that inspects tool calls, parameters and multi-step reasoning, with policy-as-code (OPA/Rego) aligned to the EU AI Act. The platform is differentiated by its EU-sovereign, on-premise deployable stack, native integrations with LangChain, Hugging Face, PyTorch, TensorFlow, MLflow, Weights & Biases, NeMo Guardrails, GitHub Actions and Pytest, and proprietary research assets (RealHarm DB, Phare benchmark with Google DeepMind, StereoTales).
The business model is a hybrid of freemium open-source (the Giskard Python library) feeding into quote-based enterprise subscriptions for Giskard Hub and Giskard Guards, plus professional services revenue from the Sofia AI security assessment (with a refund guarantee if no vulnerabilities are found). Go-to-market combines direct enterprise field sales targeting French and European regulated enterprises, product-led growth via the open-source library, and channel partnerships with Collibra and Databricks. The named customer base of 13+ enterprises is heavily concentrated in French banking (BNP Paribas, Société Générale, BPCE, Crédit Agricole, BoursoBank), insurance (AXA), retail/consumer goods (L'Oréal, Decathlon, ACCOR, Michelin, ByMyCar), healthcare (Doctolib), AI research (Google DeepMind) and government (DINUM).
Giskard firmographics
Firmographics- Name
- Giskard
- Legal name
- GISKARD AI SAS
- Website
- https://giskard.ai
- Company type
- Private
- Founded year
- 2021
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Giskard is a French AI security company that provides automated red teaming, LLM evaluation, and runtime guardrails for enterprises deploying conversational AI agents, serving regulated European customers in banking, insurance, retail, healthcare and the public sector via an open-source library and an enterprise Hub.
- Ownership category
- akta.pro rank
Giskard industry classification
Industry- Product category
- AI Security & LLM Testing Software
- NAICS
- Security Systems Services (56162), Computer Systems Design and Related Services (54151), Investigation and Security Services (5616)
- akta.pro primary industry
- Model Security Testing & Red Teaming (adversarial ML, jailbreaks) (HDAAAKAC)
- akta.pro secondary industries
- Collaboration Security for Chat & Messaging (Teams/Slack) (HDADAKAF), Penetration Testing & Red Teaming (BPAKAHAF), Secure Model Deployment & Runtime Protection (sandboxing, isolation) (HDAAAKAH), Safety & Alignment Evaluation (red-teaming, harmful capability testing) (HDAAAMAL)
Keywords
Where Giskard is headquartered
LocationHeadquarters
- HQ city
- Paris
- HQ country
- France
- HQ region
- Europe
Offices1 record
Markets served
Giskard business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Marketing or Sales, Infrastructure, Operations
Revenue model
- Giskard Hub subscription (enterprise tier): Paid SaaS subscription to the Giskard Hub, including the Continuous Red Teaming platform, LLM Evaluation platform, dashboards, RBAC, audit trails and Identity Provider integration. Pricing is not publicly disclosed; customers request a demo.
- Giskard Open-Source library (free / freemium): Free open-source Python library on GitHub used as a funnel into the paid Hub; provides scan, test suite generation, RAGET testset generation and CI/CD integration for individual practitioners and small teams.
- AI Red Teaming & Sofia assessment service: Professional services and managed engagements where Giskard's ML researchers run threat modeling, scan customer agents, prioritize findings and assist with remediation; delivered as fixed-scope assessments with a refund guarantee when no vulnerabilities are found.
- Giskard Guards (context-aware guardrails): Runtime guardrail product sold to regulated enterprises, deployable in customer infrastructure with Policy-as-Code (OPA/Rego), EU AI Act policy packs and continuous signature updates; supports on-premise and EU data residency.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Freemium | Pay-as-you-go | Giskard Open-Source library (free, self-serve) |
| Subscription | Annual | Giskard Hub (enterprise tier) — quote-based |
| Outcome Based/ Performance | Pay-as-you-go | Giskard Sofia AI security assessment |
Go-to-market motion4 records
Distribution channels5 records
Marketing channels12 records
Giskard product offering
Product offeringCore offering
Giskard provides an AI Red Teaming and LLM Security platform composed of a free open-source Python testing library and the commercial Giskard Hub enterprise platform. The Hub runs Continuous Red Teaming, LLM Evaluation and Giskard Guards (context-aware AI guardrails) against conversational AI agents via a black-box API endpoint approach, supplemented by a managed AI security assessment service staffed by Giskard's ML researchers. The offering targets regulated enterprises that need to detect hallucinations, prompt injection, data disclosure, harmful content and bias in LLM agents before production deployment.
Product overview
Giskard offers an AI Red Teaming & LLM Security platform delivered as a two-tier architecture: a free, open-source Python library (Giskard Open-Source Library, including the RAG Evaluation Toolkit / RAGET) for individual practitioners, and a commercial enterprise platform called Giskard Hub that bundles three core products — Continuous Red Teaming, LLM Evaluation, and Giskard Guards (AI Guardrails) — for organisations deploying LLM agents. These are complemented by an LLM Red Teaming service staffed by Giskard's ML researchers, plus public-facing research assets such as the Phare LLM Benchmark and the RealHarm DB. The Hub operates as a black-box testing tool that wraps any conversational AI agent accessible via an API endpoint, while Giskard Guards enforces EU-sovereign, policy-as-code guardrails on top of the same agents.
Differentiator
Problem solved
Functional benefit
Brands
- Giskard Guards: Context-aware AI guardrails product for enterprise AI agents; positioning as Europe's first independent sovereign guardrail platform.
- Giskard Hub
- Giskard Open-Source
Products and services
- Giskard Hub Enterprise AI security and quality platform that runs Continuous Red Teaming, LLM Evaluation and Giskard Guards (AI guardrails) for conversational AI agents, delivering full vulnerability reports, go/no-go deployment recommendations and remediation guidance for regulated enterprise customers.
- Giskard Open-Source Library Free Python library (pip-installable) that automatically runs exhaustive test suites via giskard.scan to identify risks in LLMs, NLP, tabular and vision models, with native wrappers for LangChain, Hugging Face, PyTorch, TensorFlow and scikit-learn, and CI/CD integration.
- Continuous Red Teaming The first Continuous Red Teaming product for LLM agents, generating dynamic multi-turn attacks via an AI red teamer, using internal business context (PDFs, knowledge bases, websites) and external threat databases (OWASP) for comprehensive attack coverage.
- LLM Evaluation Platform for building, maintaining and automating golden evaluation datasets for LLM agents, with security and quality scans, business-context testing, regression prevention, LLM-as-a-Judge customisation and alerting.
- Giskard Guards Context-aware AI guardrails for agents that inspect prompts, tool calls, parameters and multi-step reasoning, enforcing policy-as-code (OPA/Rego) and prebuilt policy packs (EU AI Act, OWASP LLM risks) to block unsafe actions in real time.
- LLM Red Teaming service (Giskard Sofia) Managed service delivered by Giskard's ML researchers that scans LLM apps for hallucinations, harmful content, prompt injection, information disclosure, robustness issues and stereotypes, then reports, mitigates and deploys with continuous monitoring; sold with a money-back guarantee when no vulnerabilities are found.
- Phare LLM Benchmark Open, multilingual LLM benchmark (Potential Harm Assessment & Risk Evaluation) covering hallucination, bias & fairness, harmfulness and jailbreak vulnerability across English, French and Spanish, developed with Google DeepMind.
- RealHarm DB Public database of real AI incidents affecting LLM applications, used by Giskard to surface the practical relevance of risks such as hallucination, which account for more than one-third of reviewed incidents.
Quantifiable outcome
- Up to 40% of blocked requests are false positives with generic guardrails (industry baseline Giskard cites)
- +5 more outcomes
Companies that use Giskard
Customer profileNamed customers14 records
Segments7 records
Ideal customer profiles3 records
Giskard technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration14 records
AI capability12 records
Feature8 records
Giskard partnerships and signals
Strategic signalPartnerships
Twelve partnerships are on record, tiered flagship go-to-market partner, ecosystem partner, flagship research partner, flagship education partner, core ecosystem partner, standard integration and core technology integration.
- Collibraflagship go-to-market partnerStrategic partnership with AI testing startup Giskard announced alongside the launch of Collibra's AI Command Center; Collibra is using Giskard's testing technology as the AI testing layer within its agentic governance platform.
- Databricksecosystem partnerCollibra (and by extension Giskard) joined Databricks' Agent Bricks ecosystem as a founding member, opening distribution of Giskard's AI testing capabilities to Databricks' agentic AI customer base.
- Google DeepMindflagship research partnerResearch and funding partner for the Phare multilingual LLM benchmark; co-developed during the Paris AI Summit to provide open, independent measurements of LLM trustworthiness across hallucination, bias, harmfulness and jailbreak resistance.
- DeepLearning.AI (Andrew Ng)flagship education partnerCo-produced the free short course "Red Teaming LLM Applications" with Andrew Ng; promoted via Giskard's blog and email list to drive awareness and inbound leads.
- GitHubcore ecosystem partnerGiskard integrates into GitHub Actions CI/CD pipelines; Giskard was also selected for GitHub's second Accelerator cohort focused on open-source AI projects.
- MLflowstandard integrationDedicated MLflow integration for LLM and tabular models, with example notebooks for Databricks.
- Weights & Biasesstandard integrationDedicated Weights & Biases integration with example notebooks for LLM and tabular use cases.
- NeMo Guardrailsstandard integrationDedicated NeMo Guardrails + Giskard scan integration for NVIDIA NeMo users.
- Hugging Facecore technology integrationNative integration of Giskard with Hugging Face transformers and the Hub; includes the Giskard Evaluator on Hugging Face and the ability to push QATestsets to the Hugging Face Hub.
- LangChaincore technology integrationGiskard wraps LangChain chains, prompts and retrieval pipelines as testable models, enabling scan and evaluation of LangChain-built LLM applications.
- PyTorchstandard integrationNative support for PyTorch models in the Giskard open-source library; logo displayed in the integrations section of the open-source page.
- TensorFlowstandard integrationNative support for TensorFlow models via a custom Giskard Model wrapper; logo displayed in integrations.
Scale indicators12 records
Recent moves6 records
Expansion highlights6 records
Giskard competitors and assessment
Company assessmentDirect peers
- Lakera: Zürich-based Lakera offers an enterprise LLM and AI agent security platform (Lakera Guard, Gandalf red-teaming game) with prompt-injection and data-leakage guardrails, directly overlapping Giskard's Hub and Giskard Guards offering for European regulated buyers.
- Protect AI: Protect AI provides an end-to-end AI security platform covering model scanning, LLM red teaming and guardrails, targeting regulated enterprises — a close functional analogue to Giskard's continuous red teaming and LLM evaluation stack.
- Arthur AI: Arthur AI sells an AI observability and security platform for production LLM and ML models, including evaluation, guardrails and drift detection, competing for the same enterprise AI governance budget as Giskard Hub.
- CalypsoAI: CalypsoAI provides an AI security and governance platform with model scanning, red teaming and policy enforcement for generative AI applications, a directly comparable enterprise offering to Giskard's Continuous Red Teaming and Guards.
- HiddenLayer: HiddenLayer offers AI security solutions for both predictive ML and generative AI, including model scanning and adversarial testing, positioning it as a competing AI security vendor for regulated enterprise deployments.
Broad incumbents
- Robust Intelligence (Cisco): Originally a direct AI firewall / red-teaming peer (now part of Cisco), Robust Intelligence brings AI model validation and guardrails into Cisco's broader security portfolio, giving hyperscaler- and platform-incumbent reach that Giskard competes against in large enterprise RFPs.
- NVIDIA (NeMo Guardrails): NVIDIA's NeMo Guardrails is an incumbent, broadly distributed runtime guardrail toolkit shipped with NVIDIA AI Enterprise; Giskard already integrates with NeMo but also competes against it as a default guardrail choice in enterprise AI stacks.
- Datadog: Datadog's LLM Observability and AI Monitoring capabilities extend its broad enterprise monitoring platform into LLM evaluation and guardrail-style use cases, making it a credible adjacent incumbent that could bundle AI security into existing enterprise contracts.
Emerging players
- Promptfoo: Open-source LLM evaluation and red-teaming framework widely used by developers; overlaps with Giskard's open-source library on testing and adversarial prompt generation, but lacks Giskard's enterprise Hub and sovereign positioning.
- Confident AI (DeepEval): DeepEval is an open-source LLM evaluation framework with a managed cloud offering, competing for the same AI/ML engineering teams that Giskard's open-source library targets.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Giskard social profiles
Digital presenceGiskard compliance and trust
Trust signalCompliance3 records
Giskard financial estimates
Financial estimateRevenue estimate
Valuation estimate
Giskard leadership team
Management profileNumber of profiles
Profiles5 records
Giskard funding detail
Funding detailFunding overview
Funding rounds6 records
Investors7 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Giskard M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Giskard
What does Giskard do?
Giskard provides an AI Red Teaming and LLM Security platform composed of a free open-source Python testing library and the commercial Giskard Hub enterprise platform. The Hub runs Continuous Red Teaming, LLM Evaluation and Giskard Guards (context-aware AI guardrails) against conversational AI agents via a black-box API endpoint approach, supplemented by a managed AI security assessment service staffed by Giskard's ML researchers. The offering targets regulated enterprises that need to detect hallucinations, prompt injection, data disclosure, harmful content and bias in LLM agents before production deployment.
Is Giskard a public or private company?
Giskard is a private company. It is classified as venture growth investor backed and is currently operating.
When was Giskard founded?
Giskard was founded in 2021. It employs 11 to 50 people.
Where is Giskard based?
Giskard is headquartered in Paris, France, in the Europe region.
How does Giskard make money?
Four revenue lines are on record. Giskard Hub subscription (enterprise tier) is the primary driver. The others are giskard Open-Source library (free / freemium), AI Red Teaming & Sofia assessment service and giskard Guards (context-aware guardrails).
Who are Giskard's main competitors?
Direct peers on record are Lakera, Protect AI, Arthur AI, CalypsoAI and HiddenLayer. Broad incumbents are Robust Intelligence (Cisco), NVIDIA (NeMo Guardrails) and Datadog. Emerging players are Promptfoo and Confident AI (DeepEval).
Does Giskard have an API?
Yes. Giskard offers APIs and a Python SDK for wrapping models (HuggingFace, LangChain, PyTorch, TensorFlow, scikit-learn, custom Python functions) and datasets, then scanning for vulnerabilities. The LLM agent under test only needs to be accessible through an API endpoint. Enterprise tier (Giskard Hub) provides programmatic API access for security/quality scans, test-suite execution, and integration with CI/CD pipelines. Developer documentation is at docs.giskard.ai.
What industry is Giskard in?
Giskard's product category is AI Security & LLM Testing Software. Its primary akta.pro industry code is HDAAAKAC, Model Security Testing & Red Teaming (adversarial ML, jailbreaks), with a secondary code of HDADAKAF, Collaboration Security for Chat & Messaging (Teams/Slack). Its NAICS code is 56162.