# One World AI

> One World AI Limited (UK company no. 17173227, London) develops OWAI OS, which lets regulated organisations run AI agents under deterministic control, on-premises or in a hybrid setup. Every AI-proposed action is checked against versioned policy, authorised by a named role and recorded as cryptographically verifiable evidence. OWAI OS is AI-agnostic: models can change while controls and evidence stay the same. Target sectors: fintech, govtech, regtech, medtech, deeptech and dual-use, plus software houses building for regulated clients. The in-house agent stack comprises OWAI Code (governed coding), OWAI Learn (upskilling staff and customers) and OWAI Evidence (research assessment). OWAI OS is in pre-production. It helps organisations prepare for assessment and certification; it does not certify systems or establish legal compliance by itself.

## Product and approach

- [OWAI OS](https://www.oneworldai.co.uk/owai-os/): The control layer for in-house AI agents — policy as code owned by compliance, a runtime decision for every proposed action, cryptographic evidence and approval by named roles; on-premises, hybrid and AI-agnostic.
- [Compliance](https://www.oneworldai.co.uk/compliance/): How OWAI supports scoped compliance work: risk context, control mapping, traceability and deployment qualification, with FCA and EU AI Act context. OWAI does not provide legal advice.
- [Assurance](https://www.oneworldai.co.uk/assurance/): What a control record establishes and what it does not; on-premises and hybrid deployment options assessed during pilot design. Technical documentation is available under NDA on request.
- [Agents](https://www.oneworldai.co.uk/applications/): The in-house agent stack under OWAI OS — OWAI Code (build software for your own processes, pre-beta), OWAI Learn (upskill staff and customers, pre-production) and OWAI Evidence (research with claims tied to evidence; research and evaluation runtime, not generally available).
- [OWAI Learn](https://www.oneworldai.co.uk/owai-learn/): Controlled runtime for AI-assisted learning with explicit learning policies and progress records.

## Research

- [Research](https://www.oneworldai.co.uk/research/): Selected preprints by Paweł Bański (CTO & Principal Architect) and Sergiusz Sobieski (CEO) on authority provenance, stopping points, execution legitimacy and enforceable AI decision boundaries, published on Zenodo with DOIs.
- [Research summary: Standing to Stop](https://www.oneworldai.co.uk/research/standing-to-stop/): Place, standing and effect — why a stopping point needs presently valid, traceable authority (DOI 10.5281/zenodo.22810377).
- [Research summary: Architectural Constraints in AI Governance](https://www.oneworldai.co.uk/research/architectural-constraints-in-ai-governance/): Execution legitimacy and trajectory-level governance (DOI 10.5281/zenodo.18980403).
- [Research summary: Enforceable Decision Boundaries](https://www.oneworldai.co.uk/research/enforceable-decision-boundaries/): Separating inference from execution authority (DOI 10.5281/zenodo.18233853).
- [Research summary: Prompt Engineering as a Control Illusion](https://www.oneworldai.co.uk/research/prompt-engineering-as-a-control-illusion/): Why prompts cannot be a control boundary (DOI 10.5281/zenodo.18213542).
- [Research summary: The Missing Place for "Stop"](https://www.oneworldai.co.uk/research/the-missing-place-for-stop/): Why systems that work correctly still fail (DOI 10.5281/zenodo.18188447).

## Insights

- [EU AI Act and AI agents](https://www.oneworldai.co.uk/insights/eu-ai-act-ai-agents-logging-human-oversight/): Articles 12, 14 and 26 for agents that act; high-risk dates after Regulation (EU) 2026/1744 (Annex III: 2 December 2027; Annex I: 2 August 2028).
- [FCA and AI agents](https://www.oneworldai.co.uk/insights/fca-ai-agents-accountability/): Accountability under SM&CR and the Consumer Duty when AI agents act; FCA applies existing rules rather than an AI-specific rulebook.

## Company and contact

- [Company](https://www.oneworldai.co.uk/company/): Sergiusz Sobieski (CEO) and Paweł Bański (CTO & Principal Architect); registered office 71–75 Shelton Street, Covent Garden, London WC2H 9JQ.
- [Contact](https://www.oneworldai.co.uk/contact/): Two options: a short briefing request (name, email, organisation, workflow) or a full pilot enquiry with KYC details; or email ssobieski@oneworldai.co.uk. Pilot process: KYC check on the organisation, first contact after a positive check, a short online meeting, then a technical discussion once a mutual NDA is signed.
- [Privacy policy](https://www.oneworldai.co.uk/privacy-policy/) · [Legal notice](https://www.oneworldai.co.uk/legal-notice/)