Giskard Hub AI Agent Testing and Red Teaming

Giskard is a French AI quality and security company co-founded by co-CEOs Alex Combessie and Jean-Marie John-Mathews, backed by investors including Elaia and supported by the European Commission's EIC Accelerator and Bpifrance. It helps organizations find security and quality failures in LLM chatbots and AI agents before and after they reach customers.

Giskard Hub continuously tests any text-based agent, from RAG chatbots to agentic workflows, through black-box access via an API endpoint. It generates risk-based test scenarios from the client's knowledge base and external security sources, detects prompt injection, data disclosure, sycophancy, hallucinations, contradictions and wrongful refusals, and offers a collaborative red-teaming playground and annotation for non-developers. An autonomous Agentic Red Teamer audits agents, runtime guardrails add protection, and audits end with a deploy-or-fix report. An open-source library is also available. Users include BNP Paribas, BPCE, Michelin and Decathlon.

• Continuous testing: tests AI agents before and after deployment via API endpoint.
• Security scanning: detects prompt injection, data disclosure, sycophancy and harmful content.
• Quality checks: finds hallucinations, contradictions, omissions and wrongful refusals.
• Test generation: builds risk-based scenarios from the client's knowledge base.
• Agentic Red Teamer: autonomous agent that audits other AI agents.
• Collaboration and guardrails: red-teaming playground, annotation tools and runtime protection.

Vietnamese banks can red-team customer chatbots for prompt injection and data leakage before launch. Insurers and telecoms can run continuous regression tests on AI agents after each model update. Public sector bodies can audit citizen-facing assistants on-premises.

Teams catch prompt injection, data leaks and hallucinations before customers or attackers find them. Black-box testing needs only an API endpoint, so it works with any model or vendor. Business experts can join red teaming and annotation without coding. Go/no-go audit reports and managed remediation give risk and compliance teams evidence for release decisions.

AI regulation such as the EU AI Act and growing incidents with customer-facing chatbots are pushing companies to test LLM systems for safety and accuracy. Banks and insurers in Southeast Asia deploying GenAI assistants face similar regulator scrutiny and need independent testing before launch.

European company identified as French on its website, co-led by Alex Combessie and Jean-Marie John-Mathews. Supported by the EU EIC Accelerator and Bpifrance; investors include Elaia and Hugging Face CTO Julien Chaumond. SOC 2 Type II and HIPAA compliant, GDPR native, with EU or US data residency and on-premise option. Partners include Hugging Face, NVIDIA Inception, Databricks, Mistral AI, Accenture and KPMG.