Governance
Policy & compliance
Havas Lynx's compliance training — the rules for using AI day-to-day, handling data safely, the EU AI Act, and AI-generated imagery.
General usage
Approved Tools Only
All AI use for client work must go through LynxAI. Free consumer tiers (ChatGPT free, Claude.ai free) are not approved. Personal AI subscriptions are for personal use only.
Human in the Loop
All AI outputs used in client work must be reviewed, edited, and validated by a qualified human. AI generates — humans are responsible. No exceptions for client-facing content.
Client Disclosure
Follow client-specific disclosure requirements (see Client Guidelines). When in doubt, disclose. Building trust on AI transparency is a competitive advantage, not a risk.
Data Handling
LynxAI is zero-retention — inputs aren't stored or used to train models. Don't input client PII or confidential data without checking the client's specific AI position first.
GDPR & Data Privacy
Processing personal data through AI requires a legal basis under GDPR. If you're unsure whether your use case involves personal data, assume it does and escalate to the Data Privacy team.
Factual Accuracy
Verify all factual claims, statistics, citations, and medical information against primary sources. AI can hallucinate confidently and incorrectly — always check.
Medical & Regulatory Content
AI-assisted medical, clinical, and regulatory content must go through formal MLR review regardless of AI involvement. AI does not replace or accelerate the medical review process.
Copyright
AI-generated content may reproduce copyrighted material. Treat all AI text and imagery as requiring the same copyright review as externally sourced content. Get legal guidance before using AI image generation for clients.
PII & data
EU AI Act
Mandatory transparency training for synthetic media on EU-distributed healthcare assets — 12 chapters, from why it matters to real-world scenarios.
AI images
Level 1 — Safest
Established & AI-Ready
In-platform tools with commercial licensing built in: Adobe Firefly, Getty AI (iStock), Microsoft Designer. No additional legal sign-off required.
Level 2 — Use with governance
Established, Needs Checks
General-purpose generators (Midjourney, DALL·E, Stable Diffusion). Requires: licensing check, disclosure label, and human creative direction to establish authorship.
Level 3 — Legal sign-off required
Emerging Territory
Patient likeness, real people's faces, clinical imagery, deepfakes. Requires legal sign-off before use — without exception.