Giskard

Researched 2026-06-28. Primary category: llm-observability; also ai-red-teaming.

One-liner — A French open-source LLM/ML testing and red-teaming platform that scans AI models and agents for quality, security, and vulnerability issues.

What it does — Giskard provides an open-source Python library to test ML and LLM applications — detecting quality regressions, robustness gaps, bias, and security vulnerabilities — plus an LLM vulnerability scanner that runs adversarial, multi-turn probes (prompt injection, jailbreaks, harmful/off-topic outputs). The commercial Giskard Hub wraps this for teams: continuous red teaming, collaborative test management, and evaluation reporting for agents. The job: systematically probe your AI before and after shipping, with evidence for risk/compliance.

Where it sits in the stack — Sits across llm-observability (the evaluation/testing sub-segment) and ai-red-teaming (adversarial probing/guardrail validation). It breaks the untrusted-input leg by surfacing how a model misbehaves under hostile prompts — a pre-production and continuous-assurance tool, not an inline runtime firewall.

Deployment & architecture — OSS library (giskard-oss on GitHub) installed as an SDK in your own environment; Giskard Hub available as SaaS or self-managed for enterprises. Integrates with common LLM frameworks and CI; partners listed across AWS/GCP/Azure and consultancies (Accenture, PwC, EY, KPMG — marketing).

Positioning & differentiators — Known as open-source-first, EU-based, and red-teaming-forward, with an explicit EU AI Act compliance angle. Closest neighbors: trulens and arize-phoenix (OSS eval/tracing), promptfoo and patronus-ai (eval + red-teaming), and dedicated red-teamers like mindgard and splxai. Differs from production-monitoring platforms (fiddler-ai, arthur-ai) by centering testing/vulnerability scanning over live observability. European data-residency and AI-Act framing are its distinctive pitch.

Ownership, funding & M&A — Independent and early-stage. Founded 2021 in Paris by Alex Combessie and Jean-Marie John-Mathews (now co-CEOs); CTO Matteo Dora. Raised a small seed (~$1.5–2M, ~2022; sources disagree on the exact figure) from Elaia and angels including founders/CTOs from Hugging Face, Mistral AI, and Uber (e.g., Charles Gorintin). Won a ~€3M Bpifrance / France 2030 grant (2024) to lead an LLM-evaluation R&D consortium with Mistral AI, Artefact, INA, and BnF. No M&A; no later equity round confirmed. (confidence: medium)

CTO / hedge-fund lens — Day-1 useful if you build LLM/agent apps and want a cheap, code-first way to red-team and test them — the OSS library is free, and the EU AI Act framing suits a fund with European entities or regulators. It is an assurance/testing tool, not a production monitor or a SR 11-7 governance system of record; pair it with a monitoring/governance layer. Early-stage vendor — diligence the Hub’s roadmap and support before depending on it.

Competitors / alternatives — promptfoo, patronus-ai, mindgard, splxai, trulens, arize-phoenix, guardrails-ai.

Open questions / to verify

• Exact seed amount and total funding (sources report $1.5M vs $2M); any round since 2022.
• Current paid-customer roster vs marketing logos (AXA, BNP Paribas, Michelin, Google DeepMind cited by vendor).

Sources

• About Giskard — fetched 2026-06-28 — supports: founders, product scope, red-teaming/EU AI Act positioning; confidence: med (vendor/marketing).
• Giskard-AI/giskard-oss (GitHub) — fetched 2026-06-28 — supports: OSS testing/eval library; confidence: high.
• Giskard — Crunchbase / Startup Intros — fetched 2026-06-28 — supports: founding 2021, Paris, seed funding, Bpifrance grant; confidence: med.

History

• [2026-06-28] Stub created from seed registry.
• [2026-06-28] Researched; confirmed independent, founded 2021 Paris, small seed (~$1.5–2M) + ~€3M Bpifrance grant. OSS testing/red-teaming + Giskard Hub. No M&A. Set ownership_confidence medium (exact funding figures conflict).