Make AI agents steerable and aligned
We engineer robust frameworks that ensure complex AI systems operate predictably and always remain under human control.
Backed by ARIA, the UK's high-risk, high-reward research agency.
Our team is featured in the ARIA TA2 Creators program, bringing expertise from Oxford, MILA, and Anthropic.
Provable Safety Through Formal Models
We develop frameworks centered on verifiable world models to ensure the safe and trustworthy deployment of AI agents. Our approach is built on the principle of keeping "humans in the loop by design" to guarantee that AI actions cohere with latent human preferences.
Our methodology provides robust safety guarantees by enabling domain experts to synthesise a formal, auditable world model and explicit safety specification. We then generate a policy for an AI agent together with a mathematical proof demonstrating that the agent, operating within this framework, will never violate the spec. By leveraging formal verification and proof assistants alongside modern large-scale machine learning stack, we reduce catastrophic risk and build systems that are robust, controllable, and faithful to human intent in complex, real-world environments.
Our Team
Younesse Kaddar
Director of Recursive Safeguarding Ltd
Oxford Univ ARIA SGAI TA1 & Opportunity Seed
Finishing PhD, formerly at MILA (worked with Yoshua Bengio)
Worked on LLM hallucinations at Cohere
Technical Cofounder and CTO of RightPick
Rob Cornish
Oxford Univ Nightingale Fellow
Cofounder and Director of Quro Medical
Moving to faculty role at NTU Singapore
Machine Learning Researcher (computational statistics, deep generative modeling, causal inference)
Sam Staton
Oxford Univ Professor
ARIA SGAI TA1 grants & Opportunity Seed (SynthStats: GFlowNet-finetuning)
ERC Consolidator Grant BLAST: Better Languages for Statistics
Pedro Amorim
Faculty at Bath Univ
Expert in programming language theory
Expert in formal verification and categorical semantics
Nikolaj Jensen
Oxford Univ PhD student in AI and theory
ARIA SGAI TA1 via Adjoint Labs Ltd
Jacek Karwowski
Oxford Univ PhD student in AI and theory
SERI MATS scholar and OpenPhil funded
Published work on probabilistic programming and RL safety
Mohammed Mahfoud
Independent researcher and entrepreneur
Led Exa-scale Safeguards Research at Anthropic
Scientist AI with Yoshua Bengio at Mila
Co-founder of Alignment team at TII (Falcon 1 LLM)
Previous EdTech startup co-founder, exited
Paolo Perrone
Oxford Univ postdoc
Projects with ARIA SGAI TA1 through Oxford and Adjoint Labs Ltd
Leading contributor to Markov category theory
Ali Zein
Tech and start-up background (Munich, Oxford, Cambridge)
Advisory board of Smith School of Enterprise and the Environment
Contact Us
We are currently operating in stealth mode while developing our core technology. For investment inquiries or to learn more, please reach out.
Send us a message
Email Us Directly
contact@recursive-safeguarding.orgWe welcome inquiries from investors, researchers, and organizations interested in AI safety. Our team will respond to serious inquiries within 48 hours.