Giskard

Giskard is a Paris-based company that builds an open-source and collaborative platform for testing, evaluating, and securing AI models, with a focus on large language models and AI agents. Its open-source Python library lets data scientists and ML engineers scan models for vulnerabilities and quality issues, generating test suites that surface hallucinations, prompt-injection weaknesses, robustness failures, harmful or biased outputs, and performance regressions. The commercial Giskard Hub extends this into an enterprise platform for team collaboration, business-domain test management, and continuous red teaming that keeps probing deployed LLM applications for new failure modes over time. Giskard positions itself around exhaustive, domain-specific testing aligned to emerging requirements such as the EU AI Act, and has collaborated with organizations including Google DeepMind and the AI Incident Database, with some work funded by the European Commission and Bpifrance. Founded in 2021 by Alex Combessie, Jean-Marie John-Mathews, and Brian Fineran (with backgrounds at Dataiku and in AI-ethics research), the company is a Y Combinator alumnus and has raised seed funding from investors including Y Combinator, Elaia, and others. Giskard is aimed at ML, quality, and risk teams that want reproducible, framework-aligned testing and continuous red teaming, delivered through an open-source-first approach they can adopt incrementally.