A UK nonprofit applying formal methods and machine learning to build open infrastructure for mathematically guaranteed AI safety assurance.
A UK nonprofit applying formal methods and machine learning to build open infrastructure for mathematically guaranteed AI safety assurance.
People
Updated 05/18/26By grantmaking.aiCo-Founder & Director
Head of Zeroth Research & Director
Co-Founder & CTO
Co-Founder & Director
Funding Details
Updated 05/18/26By grantmaking.ai- -
- -
- -
- -
Org Details
Updated 05/18/26By grantmaking.aiZeroth Research is a UK-registered nonprofit organization founded in June 2025, dedicated to formal methods and artificial intelligence research. Its mission is to build open infrastructure that provides mathematically guaranteed safety assurance for algorithmic systems, for the benefit of all.
The organization addresses two primary categories of risk in AI systems: modeling risks arising from incorrect assumptions about the digital or physical environment a system operates in, and alignment risks arising from discrepancies between a system's actual behavior and the behavior intended by designers, users, or regulators. Zeroth Research takes an integrated approach to both safety (preventing system flaws from causing harm) and security (defending systems against deliberate attacks and misuse).
Their technical framework is built on three pillars. In the pre-learning phase, they develop tools for mathematical modelling — creating auditable models of deployment environments and formal specifications of intended system behavior. In the in-learning phase, they focus on formal certification — developing technology that enables machine learning algorithms to generate not only decision-making policies but also machine-checkable proof certificates that formally witness compliance with modelled safety requirements. In the post-learning phase, they provide continual monitoring infrastructure for post-deployment validation and ongoing compliance assurance.
The founding team draws heavily from the University of Birmingham's School of Computer Science. Co-Founder and CTO Pascal Berrang is an Associate Professor in Computer Security, whose work applies cryptography and zero-knowledge proofs to AI security, privacy, and safety. Director Mirco Giacobbe is an Associate Professor whose research bridges formal methods and AI, developing automated verification techniques for cyber-physical systems. Director Luca Arnaboldi is an Assistant Professor working on neural networks and cybersecurity. Director Simon David Schmidt rounds out the leadership team.
Zeroth Research has received funding from ARIA (the UK's Advanced Research and Invention Agency) under the Safeguarded AI programme (TA1.2/1.3 grant), as well as support from FBK (Fondazione Bruno Kessler). Pascal Berrang also leads a separate ARIA-funded initiative on privacy-preserving AI safety verification and a Foresight Institute-backed project on zero-knowledge attestation for AI security.
Theory of Change
Updated 05/18/26By grantmaking.aiZeroth Research believes that empirical testing alone is insufficient to guarantee the safety of AI systems, and that formal mathematical proof is the only way to provide reliable assurance. By developing open infrastructure that combines machine learning with automated reasoning, they aim to enable AI systems to produce not just decisions but cryptographic proof certificates verifying compliance with formal safety specifications. This shifts AI safety from probabilistic/empirical assurance to mathematical certainty — analogous to safety guarantees in nuclear power or passenger aviation. Their open-infrastructure approach means these tools can be adopted broadly across safety-critical industries (avionics, automotive, medical devices, robotics), creating systemic impact on how AI systems are developed and deployed. By making safety verification accessible and rigorous, they reduce the risk of deployed AI systems causing harm through misalignment or adversarial exploitation.
Grants Received– no grants recorded
Updated 05/18/26By grantmaking.aiProjects
Updated 05/18/26By grantmaking.aiDiscussion
No comments yet. Be the first to share your thoughts.