// RESPONSIBLE AI
Responsible AI should be the baseline.
When AI supports real clients, privacy, boundaries and human accountability have to be built in from day one. Here is how Magif does it, and how responsibility is shared.
// OUR PRINCIPLES
The principles every Magif agent is built on.
Magif doesn't train its own models. It orchestrates trusted third-party models inside a secure, transparent governance framework. These principles govern how that's done.
Transparency
Every agent identifies itself as AI-powered. We disclose data flow, model providers and known limitations, and communicate significant model changes.
Human-centered design
AI augments human expertise. It never replaces it. Features support reflection, insight and productivity while the expert keeps judgment and control.
Accountability
We own the platform layer: the ethical use, control and governance of integrated AI. Experts stay accountable for outputs within their own practice.
Fairness & inclusion
We test system templates for bias and help experts validate fairness in their own use cases.
Reliability & safety
Continuous testing for stability and coherence, with incident-response and monitoring procedures for misuse or abnormal behavior.
Privacy & security
TLS 1.2+ in transit, AES-256 at rest, role-based access and MFA. Client data is never used to train external models.
// PRIVACY & DATA
Your clients' data stays in Europe, and stays yours.
Minimal, purpose-bound, encrypted and EU-based. Data only leaves our infrastructure to reach the model provider, and only after anonymisation.
Hosted in Europe
Personal information on AWS in Paris (France); expert knowledge in a Qdrant vector database in Berlin (Germany). Data stays on Magif servers by default.
GDPR by default
Minimal, purpose-bound collection. Experts can export or delete their data, and their clients’ data, at any time, with verified, logged erasure.
Anonymisation before processing
Data only leaves our infrastructure when sent to the model provider for agent responses, and passes through our anonymisation engine first, which detects PII and replaces it with tokens, then restores context on the way back.
Never used for training
Client and expert data is never used to train external models. Knowledge bases stay isolated per client and are never shared.
// OUR COMMITMENTS
The non-negotiables.
EU-hosted
Paris · Berlin
GDPR
export & delete anytime
EU AI Act
aligned & monitored
Human-in-the-loop
AI augments, never replaces
// ACCOUNTABILITY & STANDARDS
Clear roles, real oversight, documented compliance.
Under the EU AI Act, responsibility is shared and explicit. Magif provides the safe, compliant infrastructure; experts deploy it responsibly in their own practice.
Clear EU AI Act roles
Magif is the AI provider and integrator; the foundation-model provider is responsible for its own conformity; experts are AI deployers, responsible for lawful, ethical use.
Conformity measures
Risk classification of features, technical documentation, safety and bias checks, traceability of model versions and data flow, and post-market monitoring.
Human oversight enabled
Human oversight is built in for automated processes. High-risk domains (HR, education, health, finance) require deployers to add appropriate oversight.
Prohibited uses
No social scoring, surveillance, manipulative exploitation of vulnerabilities, or any unacceptable-risk activity under the EU AI Act (Regulation 2024/1689).
Governance structure
An AI Ethics & Risk Committee, a Technical Security Team, a Legal & Privacy Officer, and an Expert Relations Team oversee ethics, safety, compliance and feedback.
Transparency artifacts
Privacy policy, security overview, governance framework, data-flow white paper and model-transparency summary are available via the Trust Center on request.
Build AI the responsible way.
Start with European hosting, privacy controls and professional accountability already in place.
Policy questions: [email protected]