The Polish-led Startup Reinventing Trust in AI Governance
Cybersecurity, AI
💎 Finalist of Milipol Innovation Awards 2025, FormalFoundry.ai presents CodexScribe — a revolutionary formal verification platform reshaping AI trust in critical sectors.
At a time when artificial intelligence systems are becoming deeply embedded in national security, finance, and healthcare, one fundamental problem continues to erode trust: explainability. AI is powerful, yet its decision-making processes often resemble black boxes—opaque, non-verifiable, and susceptible to error. In mission-critical scenarios where mistakes can carry catastrophic consequences, this opacity is no longer acceptable.
FormalFoundry.ai, a startup co-founded by Polish entrepreneur Grzegorz Kunicki and based partly in Poland, has emerged as a finalist in the prestigious Milipol Innovation Awards 2025 in the category of Cybersecurity and AI. Their flagship platform, CodexScribe, aims to rebuild trust in AI by eliminating ambiguity and ensuring mathematical traceability across AI decisions. The team will showcase their solution at Milipol Paris 2025 (stand 4VD24, Milipol LAB) from November 18–21.
Subscribe to Startup Digest to stay ahead with the latest news, investments, and must-attend events.
CodexScribe: From Policy to Proof
CodexScribe is not just another AI governance tool. Rather than layering interpretability filters onto existing models, it shifts the foundation altogether — embedding trust through formal logic and structured domain modeling. In essence, CodexScribe allows organizations to define policies, roles, resources, and rules in formalized proof languages, creating a fully auditable and verifiable “model of the world.”
Take for instance a compliance directive: “A data analyst must not share personal data outside the EU unless explicitly authorized by vendors A or B under contract XYZ.” In CodexScribe, this instruction becomes a machine-readable, mathematically provable rule — understandable to both legal teams and computer systems. This is achieved by converting human intent into formal specifications using specialized “proof assistant” languages previously reserved for academia or military-grade applications.
Replacing Black Boxes with Transparent Infrastructure
Traditional AI models are built with statistical learning and neural approximations. While effective, these systems lack guarantees. CodexScribe introduces a specification layer that governs what AI systems can or cannot do — like a constitution enforced at code level. This architecture offers:
Internal consistency across all decision-making rules
Contextual modeling of roles, exceptions, permissions, and obligations
Automatic reasoning that allows AI to simulate scenarios and assess policy compliance before taking action
Audit trails for regulators and security teams to validate that systems are behaving as designed
This is especially vital in sectors such as defense, intelligence, and fintech where policy compliance must be provable—not just probable.
Strategic Collaborations and National Relevance
FormalFoundry.ai is not working in isolation. In 2024, the team partnered with e-Orzecznik, a Polish med-tech startup digitizing the occupational health sector. Their joint project resulted in a “hallucination-free” AI assistant for doctors and administrators—each response is grounded in formal logic and supported by mathematically verifiable rules. The system’s ability to ensure 100% reliability in regulatory compliance represents a new benchmark for health AI tools.
Moreover, the CodexScribe platform is designed to handle multiple levels of abstraction. It supports modeling of abstract organizational roles (e.g., auditors, operators), precise data types (e.g., location data, financial records), and conditional logic (e.g., “only on EU territory,” or “with supervisor approval”). The result is a seamless bridge between regulatory language and executable software behavior.
Poland’s Deep Tech on the Global Stage
The success of FormalFoundry.ai is emblematic of Poland’s emergence as a hub for deep tech innovation. Co-founder Grzegorz Kunicki brings a background in computational linguistics and systems thinking, while the international team unites expertise from high-assurance software, AI safety, and operational security. Their nomination at Milipol highlights that Poland is not only a consumer of defense technology, but a creator of the critical software infrastructure shaping future AI regulation.
In a digital world increasingly reliant on autonomous systems, CodexScribe offers what others don’t: provable reliability. Instead of “trust us,” organizations can now say, “here’s the math.”
Implications for AI Policy and EU Regulation
With the EU AI Act nearing enforcement and regulatory scrutiny rising globally, tools like CodexScribe could become a requirement rather than a choice. For regulators, it offers a direct mechanism to validate algorithmic behavior against legal frameworks. For institutions, it enables airtight compliance and transparency.
Milipol’s recognition of FormalFoundry.ai underscores the urgent demand for such tools. As the only finalist in the cybersecurity & AI category using formal methods at its core, FormalFoundry.ai stands out for its principled, non-black-box approach to AI safety and alignment.
Meet FormalFoundry.ai at Milipol Paris 2025
The team invites visitors to experience CodexScribe in action at Milipol LAB (stand 4VD24), where they will demonstrate how AI can be designed from the ground up to earn trust, not just demand it.
Tags: AI, Cybersecurity
