Signet: AI Assurance for Production Systems - Cambridge Machine Learning

Signet: Assurance for AI systems in production

A joined-up assurance layer for your AI: monitoring, observability, evaluation, safety and EU AI Act compliance - designed for systems already in use by real users.

Quick assurance assessment (1 - 3 days) Monitoring & observability Safety & guardrails EU AI Act & governance

A single, coherent AI assurance layer

Signet helps you keep your AI systems safe, reliable and auditable in production. We bring together monitoring, evaluation, safety, compliance and governance into one simple assurance flow.

We start with a fast, focused assessment to understand your current state. From there, we design and implement the level of assurance you actually need - from lightweight telemetry through to EU AI Act alignment.

See how your AI is really behaving in production
Detect drift, failures and risky behaviour early
Build clear, defensible evidence for customers and regulators

Align to the EU AI Act without over-engineering
Give product, risk and legal teams a shared view
Turn AI assurance into a strategic advantage, not a blocker

Signet assessment demo

Example output from a Signet assessment, using anonymised data across 186 checks grouped into themes such as Bias, Safety, Monitoring, Compliance and Governance.

75%

overall alignment score

Strongest themes: Governance, Security, Bias.

Key opportunities: Legal, Environmental & Social Responsibility.

Scores by assurance category

Click a bar to see details for that category (average score and number of checks).

Highest scoring areas

Showing top checks across all themes.

Key improvement areas

Showing key opportunities across all themes.

The Signet AI Assurance Suite

Foundation Assessment

In 1 - 3 days, get a concise assurance baseline and a prioritised action plan for your live AI systems.

System & data mapping
Risk & gap analysis across key themes
Concrete actions, owners and timelines

Monitoring & Observability

Make your AI observable: see what it’s doing, when it breaks, and how behaviour shifts over time.

Production telemetry & interaction logging
Quality, drift and anomaly monitoring
Incident investigation & audit trails

Safety & Evaluation

Reduce harmful or low-quality outputs with structured tests, guardrails and human oversight where needed.

Red-teaming & safety evaluations
Guardrails, policies & fallback design
Automated regression tests for prompts & models

Compliance & Governance

Turn regulations and standards into clear, manageable controls and documentation.

EU AI Act risk & role classification
Controls aligned to GDPR, ISO/IEC 42001, NIST AI RMF
Evidence packs, runbooks & team training

EU AI Act - part of a bigger assurance picture

The EU AI Act is now in force, with obligations phased in over the coming years. If your AI system is offered in or used by users in the EU, you’ll need to align with its risk-based requirements. Non-compliance can lead to fines of up to €35m or 7% of global turnover.

Signet treats the EU AI Act as one piece of a broader assurance strategy: the same monitoring, evaluation and governance that make your AI safer also make you more compliant.

Risk classification & in/out of scope analysis
Provider vs deployer duties clarified
Transparency & AI user notices

Data & evaluation obligations mapped
Monitoring, logging & incident handling
Governance, records & audit-ready evidence

CML helps you implement pragmatic, right-sized controls so you stay compliant without slowing product teams down.

Typical engagement flow

Discovery (90 mins): map system, data, users, risks.
Assessment (1-3 days): concise report & action plan.
Implementation (2-12 weeks): monitoring, safety, controls and runbooks.
Assurance (quarterly): reviews, metrics, updates and new risks.

Timeline varies by scope and criticality. We work with funded startups through to enterprises.

Why CML

Real-world production experience across financial services, retail, health and manufacturing
End-to-end expertise across LLMs, telemetry, evaluation, safety, compliance and governance
Lightweight, standards-aware approach (ISO/IEC 42001, NIST AI RMF, UK assurance guidance)
Clear, executive-friendly documentation for investors, customers and audits

Speak to an expert