Algorithm Audit provides a public knowledge platform and open‑source toolkit that helps organizations identify AI systems, assess risk, and demonstrate compliance with the EU AI Act, GDPR, and emerging standards. Its AI AQT questionnaire engine, bias detection, synthetic data generation, and generative AI evaluation tools integrate via APIs into governance dashboards, while independent audit teams deliver sociotechnical compliance reports.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations that develop or deploy algorithmic systems often lack a systematic method to identify, classify, and demonstrate compliance with the EU AI Act, GDPR, and emerging AI standards. This creates legal exposure, opaque risk management, and difficulty in detecting bias or non‑discriminatory outcomes in high‑impact AI applications.
Solution
Algorithm Audit operates a public knowledge platform that bridges policy, academic research, and practical experience with a suite of open‑source technical tools. The AI and Algorithms Qualification Toolkit (AI AQT) translates complex legal definitions into dynamic questionnaires that pinpoint in‑scope AI systems and assign the appropriate risk category. Complementary tools provide unsupervised bias detection, synthetic data generation, and a validation framework for generative AI (LLMs). Expert audit teams produce independent sociotechnical assessments, legal compliance reports, and standards contributions, which are published in a searchable knowledge base. The platform’s APIs enable seamless integration of audit outcomes into existing governance dashboards and risk registers, allowing public‑sector bodies and private firms to implement AI governance at scale.
Target Audience
Primary customers are public‑sector agencies (municipalities, ministries, EU bodies) and private enterprises that operate high‑impact AI systems and require demonstrable compliance with the AI Act, GDPR, and sector‑specific AI standards.
Features
- AI AQT: open‑source questionnaire engine that maps AI Act and GDPR criteria to risk categories, supporting organization‑wide identification of AI systems.
- Unsupervised bias detection tool: statistical hypothesis‑testing pipeline for quantitative bias analysis across datasets and model outputs.
- Synthetic data generation tool: privacy‑preserving data synthesis for testing, validation, and risk assessment without exposing real personal data.
- Generative AI evaluation framework: validation of large language model guardrails, content filters, embedding strategies, and user‑interaction designs, released under a CC‑BY‑4.0 license.
- Modular API: export of audit results, risk classifications, and compliance artefacts to governance dashboards, risk registers, and third‑party compliance management systems.
- Public knowledge base: curated articles, white papers, implementation guides, and standards documentation searchable by topic and jurisdiction.
- Standards participation: active contributions to CEN/CENELEC JTC21 working groups, influencing harmonised AI Act standards for risk management, transparency, and human oversight.
- Independent audit reports: detailed sociotechnical and legal assessments that include actionable mitigation recommendations and are made publicly available.