AI · AI Security · Research
Intelligent systems,secured by design.
Horizon Mind is an AI research and product company. We build intelligent systems — and we work on the security of the systems themselves: the models, the agents, and the infrastructure they run on.
- Research
- Adversarial robustness, privacy-preserving inference, interpretability.
- Security
- Threat models for the models, the agents and the pipelines behind them.
- Engineering
- Robust, observable systems that hold up once they ship.
What we build
Four areas of work
Each of these exists to support the others: systems that are intelligent, and secure because of how they are built.
- 01
AI
Intelligent applications and the models that power them, built to run in production rather than in a notebook.
- 02
AI Security
Protection, privacy, monitoring and governance for AI systems — treated as part of the architecture, not a layer added afterwards.
- 03
AI Agents
Agentic systems that carry out real work under clear constraints, with observable behaviour and defined boundaries.
- 04
Intelligent Infrastructure
The systems and infrastructure secure AI applications depend on, from data handling through to deployment.
Research
Research is where the products start
We work the path from an open question through to something other people can rely on. Published work appears here as it becomes available.
Research → Paper → Prototype → Product
- 01
Research
A question the literature does not settle — usually about how a system behaves under pressure rather than on a benchmark.
- 02
Paper
Method, results and limitations written down, so the work can be checked instead of taken on trust.
- 03
Prototype
The narrowest build that shows the result still holds outside the conditions it was found in.
- 04
Product
Hardened, documented and supported — the point at which someone else can depend on it.
Current directions
- Security properties of agentic systems
- Evaluation of model behaviour under adversarial input
- Privacy-preserving inference
- Monitoring and interpretability for deployed models
Technology
The ground we work on
The capability areas our research and our products are built on.
Where AI systems are exposed
- Training data and model weights
- Prompt and tool-call boundaries
- Retrieval and context sources
- Agent permissions and scope
- Inference endpoints and credentials
- Downstream handling of model output
01
Artificial Intelligence & Machine Learning
Model development, evaluation and deployment, with the evaluation treated as seriously as the training.
02
Computer Vision
Visual understanding systems: detection, recognition, tracking and inspection under real-world conditions.
03
LLM / VLM
Language and vision-language systems applied to reasoning, retrieval and document understanding.
04
AI Agents
Tool-using systems with explicit scope, permissions and audit trails.
05
AI Security
Threat modelling for AI systems: model and data integrity, prompt and input attack surfaces, access control and runtime monitoring.
06
Secure AI Infrastructure
Data pipelines, serving infrastructure and deployment paths designed around least privilege and observability.
Why Horizon Mind
How we work
Five commitments that decide what we take on and how we build it.
Research-driven
We turn research into practical technology.
Security-first
Security is part of the system, not an afterthought.
Engineering-focused
We care about robust, scalable, production-ready systems.
Product-minded
We build technology that solves real problems.
Long-term thinking
We develop systems designed for an evolving AI landscape.
Solutions
Applied to your problem
We take on work where the difficulty is technical — where the answer is not known in advance and the system has to hold up once it ships.
AI Product Development
Taking an AI product from problem definition through to a system that runs in production.
AI/ML Engineering
Model development, evaluation, and the engineering around it that makes results reproducible.
AI Security
Threat modelling, review and hardening for systems that depend on models and agents.
Computer Vision
Vision systems for inspection, monitoring and automation.
AI Agent Development
Agentic systems scoped to a real task, with the constraints designed in.
Work with us
Let's build what comes next.
If you are working on something where intelligence and security both matter, we would like to hear about it.
Start with a topic