Principled Intelligence

Spectral is an automated testing and evaluation platform that stress-tests AI agents before and after deployment. It runs thousands of scenario-based tests across threat categories, including adversarial inputs, edge cases, and long-horizon conversations, surfacing failure modes before they reach real users. Each scenario is scored, classified, and traced, generating actionable artefacts for every stakeholder: ranked failure backlogs for engineering teams, regulator-ready evidence packets for compliance functions, and portfolio-wide risk views for executives. Spectral integrates into CI/CD pipelines as a pre-release gate, runs on scheduled cadences against live systems, maps findings to frameworks such as the EU AI Act, and supports post-incident reproduction and patch verification. It closes the gap between the speed of AI deployment and the pace of institutional oversight.

Orbitals is an open-source library of AI guardrails designed to make responsible AI deployment accessible to every development team. It intercepts and classifies inputs and outputs in real time, enforcing operational boundaries on AI systems before they produce unintended or unsafe behaviour. The library is powered by specialised small language models developed by Principled Intelligence, purpose-built to deliver state-of-the-art classification at a fraction of the cost and latency of frontier models. Its first model, ScopeGuard, determines whether user requests fall within a defined operational scope, returning an explainable log of each decision. Designed for sub-second inference and multilingual coverage, it is built to run as a production guardrail on live AI traffic. Orbitals is available on GitHub and Hugging Face. By releasing it to the open ecosystem, Principled Intelligence lowers the barrier to safe AI deployment and establishes a shared, transparent foundation for AI governance tooling across Europe and beyond.