EU AI Act

• Data governance (Art. 10) — Training, validation, and operational data must be subject to data governance and management practices. Tokenized inputs are easier to govern than free-form prompts.

• Record-keeping (Art. 12) — High-risk AI systems must enable automatic logging. Cypherz logs every request, action, and metadata field.

• Transparency (Art. 13) — Users must be told they're interacting with AI. Cypherz doesn't generate the disclosure but documents which requests touched which models.

• Human oversight (Art. 14) — Operators must be able to intervene. Cypherz's dashboard surfaces every request so you can audit, alert, and revoke keys in real time.

• Audit-grade logs — Every tokenize, detokenize, and proxy call is recorded with timestamps, action types, and projection metadata — SIEM-ready.

• Project isolation — Per-project encryption keys make it straightforward to scope data flows to specific products or features.

• Provider-portable — Switch from OpenAI to Anthropic to a local model without re-architecting your data protection layer.

The Act is being phased in. Conformity assessments, registration, and certification obligations vary by risk class. Get specialist advice.