sci2sci Raises €1.2M to Trace AI Claims

The big picture: Berlin-based startup sci2sci has secured €1.2 million in pre-seed funding to develop technology that traces AI-generated claims back to their original data, preventing hallucinations in biopharmaceutical applications.

Why it matters:

• AI risks: The funding round follows recent incidents where OpenAI and Anthropic models accessed real companies without permission, highlighting critical process design flaws in AI systems.
• Regulatory pressure: The FDA has issued warnings regarding AI misuse in drug manufacturing, and the upcoming EU AI Act imposes significant fines for non-compliant high-risk AI systems, creating a clear market need for robust AI governance.
• Preventative approach: While competitors focus on detecting AI errors post-deployment, sci2sci aims to prevent unsupported claims from occurring at all, providing an auditable trail compliant with FDA standards.

How it works:

• Integrity Cortex: sci2sci’s main product extracts facts from company documents and stores them in a formal logic system, requiring each claim to be supported by direct quotes and verified by a symbolic engine.
• Hallucination detection: The system identifies invented facts and flags any related conclusions, ensuring that AI outputs are grounded in verifiable data.
• VectorCat & Parseltongue: The company’s second product, VectorCat, acts as a data catalog, linking various data sources without moving them, utilizing the open-source Parseltongue framework.

The catch: The AI governance market is highly competitive, with companies like Pharosyn, Rivia, and White Circle also securing significant funding for AI traceability and error detection. Sci2sci’s expansion into the banking sector, a highly regulated domain with complex, country-specific compliance requirements, will test the adaptability of its FDA-compliant system before its next funding round.