Giskard AI
Giskard provides a trust layer for AI agents: continuous testing, evaluation and red teaming to find hallucinations, security vulnerabilities and business-alignment failures before and after production. Its goal is to help companies deploy AI that can be trusted by its makers and users.
At a Glance
- Enterprise AI and machine-learning teams
- Data scientists and ML engineers
- AI security and red-team teams
- Product and domain experts
- +3 more
AI Tools by Giskard AI
(2)
Giskard Open Source
Python Library for Testing AI Agents
Giskard
LLM Security Testing Platform
Discussions
No discussions yet
Be the first to start a discussion about Giskard AI
Latest News
Giskard Hub 3.0: red team any AI agent, whatever its API
Agentic Red Teamer automated red-teaming benchmarked against Garak, Promptfoo and Claude Code
Launched Giskard Guards, an EU-sovereign context-aware guardrail platform
Announced Giskard Open Source v3 for agentic evaluation and red teaming
Products & Services
Enterprise platform for collaborative LLM/AI-agent evaluation and continuous red teaming, with a business-user UI, developer SDK, shared workspaces, annotation, RBAC, automated scanning, reporting and monitoring.
Apache-2.0 Python library for programmatic testing and evaluation of AI and agentic systems. The current v3 architecture includes giskard-checks for evals and giskard-scan for vulnerability, red-team and RAG/quality scanning.
Context-aware runtime control layer for prompts, tool calls and responses, using Policy-as-Code, customizable detectors and EU AI Act/OWASP compliance packs; designed for on-premise and regulated environments.
Automated agentic red-teaming assessment that dynamically attacks AI agents, produces an audit report and reusable attack dataset, and was benchmarked against Garak, Promptfoo and Claude Code.
Market Position
Giskard positions itself as an AI-agent trust layer combining quality evaluation, automated security red teaming, business-alignment testing and runtime guardrails. It differentiates from developer observability/evaluation platforms such as LangSmith by proactively generating domain-specific tests and detecting security flaws and hallucinations, and from security scanners such as Garak and Promptfoo through deeper dynamic, multi-turn and context-aware assessment; Giskard publicly benchmarked Agentic Red Teamer against Garak, Promptfoo and Claude Code.
Leadership
Founders
Alex Combessie
Co-founder and co-CEO. He met Jean-Marie John-Mathews at Capgemini in 2015, later returned to school to become a data scientist, and subsequently became a software engineer; the company describes him and the other founders as having worked on business-critical AI systems at Dataiku and Thales.
Jean-Marie John-Mathews, PhD
Co-founder and co-CEO (formerly CPO). He is an AI-ethics researcher and data scientist who met Alex Combessie at Capgemini in 2015; Giskard says the founders previously worked on business-critical AI systems at Dataiku and Thales.
Andrei (Andrey) Avtomonov
Co-founder and former CTO. He joined Alex Combessie and Jean-Marie John-Mathews after engineering/ML work and is described by Giskard as one of the founders with previous experience on business-critical AI systems at Dataiku and Thales.
Executive Team
Alex Combessie
Co-founder & co-CEO
Former strategy consultant who became a data scientist and software engineer; met Jean-Marie John-Mathews at Capgemini and helped found Giskard.
Jean-Marie John-Mathews, PhD
Co-founder & co-CEO
Data scientist and AI-ethics researcher; previously worked with Alex at Capgemini and on business-critical AI systems at Dataiku and Thales.
Board of Directors
Founding Story
Alex Combessie and Jean-Marie John-Mathews reconnected over drinks on the Paris canals after first meeting at Capgemini. Their experience building business-critical AI systems, combined with rising concerns about model quality, bias, reliability and safety, led them to create a quality-assurance platform for AI models; Andrei Avtomonov subsequently joined as co-founder and CTO. The name references Isaac Asimov's robot Giskard and the fictional 'zeroth law' about preventing harm to humanity.
Business Model
Revenue Model
Enterprise software and services model centered on paid Giskard Hub/Guards deployments, enterprise subscriptions or contracts, SDK/API access, on-premise deployments and AI audit/red-teaming services; Giskard Open Source is an Apache-2.0 community edition and upgrade path to Hub.
Pricing Tiers
The public site directs prospects to request pricing/demo; a free limited Agentic Red Teamer offer is advertised for startups and small teams.
Target Markets
- Enterprise AI and machine-learning teams
- Data scientists and ML engineers
- AI security and red-team teams
- Product and domain experts
- Regulated industries and public sector
- Financial services, banking, insurance, healthcare and customer-service organizations
- Pre-production testing and release gating for LLM applications and AI agents
- Continuous post-deployment monitoring and red teaming
- RAG quality, groundedness and knowledge-base evaluation
- Customer-service, banking and other regulated conversational assistants
- Testing classification, extraction, scoring and other non-chat JSON APIs
- AI security assessment for prompt injection, jailbreaks, data leakage and harmful behavior
- AXA
- BNP Paribas
- Michelin
- Société Générale