Stanhope AI

Stanhope AI is a London-based deep-tech company applying Active Inference and the Free Energy Principle — frameworks originating in computational neuroscience — to engineering systems for autonomous machines. Its software teaches robots and embodied platforms to make decisions in the real world, including in situations for which they have received no training, by building lightweight world-model representations rather than relying on large static training datasets. The company positions this as distinct from deep learning and from large language models, arguing that its agents possess agency: they act in order to understand their environment, combining information-seeking with task performance by minimising Free Energy.

The company's Free Energy Minimizing (FEM) world model was demonstrated in drones, where the model controlled drone behaviour, enabled flexible mission planning and human-machine teaming, and allowed the agent to describe its own reasoning and decision-making across three distinct real-world missions. A version referred to as FEMalpha featured in these demonstrations, and the company held an Active Inference exhibition and demo event in September 2024. Its more recent framing is a "Real World Model" — described as a framework for adaptive intelligence designed for dynamic physical environments beyond the limits of LLMs.

Stanhope AI's models are designed to run on-device and at the edge, using minimal data and energy, which the company presents as reducing cost and energy usage and enabling operation where communications with operators are unavailable, such as electronic warfare environments. Target deployment areas cited include autonomous systems, defence technology, industrial automation and embedded devices.

Founding story

Stanhope AI grew out of neuroscience research at University College London, where Professor Rosalyn Moran — who came to the field from electrical engineering — worked with Professor Karl Friston, whose research was being formalised into the Free Energy Principle. The two saw an opportunity to spin their ideas out as an alternative approach to physical-world AI. Moran founded the company after roughly 20 years of research as a vehicle to apply her findings to real-world situations. The company's own About page states it was founded in 2021, while its 2026 seed press release and subsequent coverage describe it as founded in 2023 and as a spin-out from University College London and King's College London.

Business model

Stanhope AI develops AI decision-making software for autonomous machines, intended for use by robotics, drone and other hardware makers; its site invites organisations to contact it about using the FEM model in their autonomous machines, and it integrates with traditional computer vision stacks.

Traction

Stanhope AI's technology is being tested in autonomous drone and robotics applications with international partners and with unnamed robotics and drone companies. Demonstrations show drones powered by Stanhope agents navigating new environments with obstacles. The CEO stated the company has moved from foundational research and early prototypes to production-grade systems operating in real customer environments.

Latest developments

On 12 February 2026 Stanhope AI announced an $8 million seed round led by Frontline Ventures, with Paladin Capital Group and Auxxo Female Catalyst Fund participating and follow-on investment from UCL Technology Fund and MMC Ventures. The funding supports its 'Real World Model', expanded deployments, team growth and field trials in 2026.