Article 50 Transparency Checker

The Article 50 Transparency Checker works out which of the EU AI Act's four transparency duties apply to each AI system you provide or deploy — AI-interaction disclosure (Art. 50(1)), synthetic-content marking (Art. 50(2)), emotion-recognition and biometric-categorisation notice (Art. 50(3)), and deep-fake and public-interest-text disclosure (Art. 50(4)). You build a small register of systems, set your role for each, and answer the duty questions; every question carries its article reference and "unsure" is treated conservatively and flagged. Each conclusion is the deterministic product of the recorded answers, and the full answer trail is shown behind it — so a reviewer holding the official text can re-derive every outcome. The tool applies the right deadline (2 August 2026, or the 2 December 2026 legacy-marking transitional) and rates residual risk from readiness and exposure, forcing the rating to high once a duty is in force and unmet. It covers Article 50 only — Article 5 prohibitions, the high-risk regime and GPAI obligations are out of scope — and runs entirely in the browser: nothing you enter leaves the page. Export the register to Excel or print the assessment memo.

• A register of AI systems, each with a provider / deployer / both role that decides which duties are yours
• The four Article 50 duties assessed by guided questions, each carrying its article reference
• A reperformable trail — every conclusion traceable to the answers and the cited provision
• The applicable deadline per duty, including the 2 December 2026 legacy-marking transitional
• A documented residual-risk rating from readiness × exposure, forced high once a duty is in force and unmet
• Excel register export, a printable assessment memo, and JSON save / load
• Fully browser-based — nothing you enter leaves the page

The Digital Omnibus deferred the EU AI Act's high-risk regime, and a lot of teams read that as breathing room. Article 50 was not deferred: its transparency duties for new systems apply from 2 August 2026, backed by fines up to €15 million or 3% of turnover. The hard part is that the duties are conditional — role, intended use, exemptions — so "does this apply to us?" has to be answered system by system, and the answer has to hold up. This tool turns that into a structured, reperformable assessment: a register of systems, guided duty questions anchored to the article, a determination with its full trail, the deadline, and a residual-risk rating — exportable for the file. It complements the AI Act Compliance Clock, which times the whole regime, by going deep on Article 50 alone.