# How does Cyber Crucible relate to the EU AI Act?

**Short answer:** Cyber Crucible uses AI only during development (its Genetic AI discovery work) and runs fixed, deterministic heuristics at runtime — there is no live AI model making decisions on the customer's endpoint. So Cyber Crucible does not place a high-risk or general-purpose AI system into the customer's environment. The customer remains responsible for assessing its own obligations under the EU AI Act, whose requirements are phasing in through 2026–2027.

## Why the architecture matters here

- **No runtime AI system deployed to the customer.** The endpoint runs deterministic kernel heuristics, not a live model — so there is no on-device AI system for the customer to classify and govern under the Act.
- **Development-time only.** The AI work happens in Cyber Crucible's internal development environment, on internal research datasets; no customer content, credentials, or keys are used to train, feed, or tune a model.
- **Transparency.** Because runtime behavior is deterministic and testable rather than probabilistic, it is straightforward to describe and document.

## The honest boundary

Cyber Crucible does not claim the product is "EU AI Act compliant," and it does not make that determination for a customer. This page describes how the product works so a customer's own AI Act assessment can account for it accurately. This is not legal advice.