AI Safety Evaluation, Red Teaming & LLM Security Testing Expertise

About Giskard

Giskard is a European AI security company founded in 2021, headquartered in Paris, France. We specialize in the evaluation, auditing, and security testing of frontier Large Language Models (LLMs) and AI agents. Our mission is to ensure that AI systems deployed in critical applications are trustworthy, reliable, and aligned with security and regulatory standards.

Core Research & Technical Expertise

• LLM Red Teaming: Continuous generation of sophisticated adversarial attack scenarios to proactively identify vulnerabilities in LLM agents before deployment.

• Hallucination Detection & Factuality Assessment: Domain-specific evaluation of AI outputs against verified knowledge sources, including RAG-based system evaluation.

• Multilingual LLM Benchmarking: Design and operation of the Phare benchmark (Potential Harm Assessment & Risk Evaluation), a multilingual evaluation framework covering hallucination, bias, harmful content, and adversarial robustness across English, French, and Spanish.

• Bias & Fairness Evaluation: Systematic measurement of discriminatory outputs and stereotype reinforcement in model responses.

• Security & Adversarial Robustness Testing: Assessment of model resistance to prompt injection, jailbreaking, and other adversarial manipulation techniques.

Proven Track Record in Publicly Funded Research Consortia

1. ArGiMi Consortium — France 2030 / Bpifrance (2024): Giskard leads the evaluation and safety workstream within the ArGiMi consortium, comprising Artefact, INA, BnF and Mistral AI.

2. European Commission — EIC Accelerator Horizon 2030

3. European Commission — LLMs4EU project: Giskard is leading evaluation of 2 AI applications, in healthcare and for public service

Partnership Interest

We are seeking collaborative partnerships in European-funded research projects focused on AI safety, trustworthy AI, LLM evaluation methodologies, adversarial robustness, multilingual AI assessment, and AI regulatory compliance (EU AI Act). We bring deep technical expertise in automated AI testing, a proven track record in publicly funded consortia, and strong connections to the European AI safety research ecosystem.