The archive · AI & Models · Strategic decision · 2021–2026
Giskard bets EU AI Act makes model testing a legal duty; 5.5k GitHub stars
Paris startup turns an open-source LLM test library into AI Act compliance evidence; €1.5M seed, €3M EU/France grant, Banque de France and L'Oréal customers.
Giskard
What the business is
Giskard sells AI quality assurance: an open-source Python library and SaaS platform that scan LLMs and AI agents for bias, hallucination, prompt-injection and safety issues, generate test suites, and produce evidence usable for EU AI Act compliance.
Starting capital：€1.5M seed led by Elaia with Bessemer Venture Partners and notable angels (Dec 2022); €3M grant from Bpifrance and the European Commission's EIC Accelerator (2023).
How it started
Founded in Paris in 2021 by ex-Dataiku and Thales engineers Alex Combessie, Jean-Marie John-Mathews and Andrei Avtomonov, who saw testing done ad hoc while AI incidents mounted (2,000+ reported since 2019); they raised a €1.5M seed led by Elaia in December 2022.
What happened
Giskard passed 1,000 GitHub stars and won a €3M grant from Bpifrance and the European Commission's EIC Accelerator in July 2023 to build SaaS for the forthcoming EU AI Act. TechCrunch profiled the framework in November 2023, when the ~20-person startup was already selling its AI Quality Hub to Banque de France and L'Oréal. The AI Act entered into force on August 1, 2024, its prohibitions applied from February 2, 2025, and general-purpose AI obligations from August 2, 2025 — validating the bet that testing would become legal evidence; the open-source project grew to roughly 5,500 GitHub stars by July 2026.
How it ended up
Still running: the framework is a leading open-source LLM and agent testing/red-teaming library, and Giskard is scaling its enterprise platform for GenAI customers across Europe.
Background
Giskard is a Paris-based startup selling AI quality assurance: an open-source Python library and SaaS platform that scan LLMs and AI agents for bias, hallucination, prompt injection and safety issues, generate test suites, and produce evidence usable for compliance. Its bet was that the EU AI Act would make model testing and documentation a legal duty, turning its developer tool into the entry point for an enterprise compliance platform.
Founded in 2021 by ex-Dataiku and Thales engineers Alex Combessie, Jean-Marie John-Mathews and Andrei Avtomonov, Giskard raised a €1.5M seed led by Elaia in December 2022. By July 2023 it had passed 1,000 GitHub stars and won a €3M grant from Bpifrance and the European Commission's EIC Accelerator explicitly to build SaaS for the forthcoming EU AI Act. TechCrunch profiled the framework in November 2023, when the ~20-person startup was selling its AI Quality Hub to Banque de France and L'Oréal.
The AI Act entered into force on August 1, 2024, prohibitions applied from February 2, 2025, and general-purpose AI obligations from August 2, 2025, validating the bet that testing would become compliance evidence rather than best practice. As of September 2026 Giskard is still running and scaling its enterprise platform for GenAI customers, with the open-source project at roughly 5,500 GitHub stars.
What has to be true
- The regulatory opening is dated and concrete: AI Act entry into force Aug 1, 2024, prohibitions Feb 2, 2025, GPAI obligations Aug 2, 2025.
- Giskard publicly built for it early: the 2023 Bpifrance/EIC grant explicitly funded EU AI Act compliance SaaS, a year before the rules applied.
- Traction is specific: €1.5M seed (Dec 2022), 1,000-star milestone (Jul 2023), TechCrunch profile (Nov 2023), ~5,500 GitHub stars (Jul 2026), named customers Banque de France and L'Oréal.
- The wedge is clear: an open-source testing library owned the developer workflow first, so regulation-converted demand flows into its paid hub.
What can be applied
When regulation turns a good practice into a legal duty, the open-source tool that already owns the workflow becomes the default vendor — the compliance deadline does the selling.
Aftermath
As of 2026-09-02 Giskard is live and scaling: the open-source giskard library (roughly 5,500 GitHub stars) tests and red-teams LLMs and AI agents, while the commercial AI Quality Hub serves enterprise GenAI customers across Europe. The EU AI Act's phased obligations — prohibitions from February 2025, general-purpose AI duties from August 2025, and high-risk system rules from August 2026 — continue to expand the market for testing and documentation tools, which is exactly the demand Giskard positioned itself to capture.
Sources
- Helping keep quality in AI industry job one, Giskard raises €1.5 million
- 1,000 GitHub stars, 3M€, and new LLM scan feature
- Giskard's open source framework evaluates AI models before they're pushed into production
- Implementation Timeline — EU AI Act
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card