EU AI Act transparency rules take effect as enforcement procedures expand

The EU AI Act’s Article 50 transparency obligations took effect on 2 August 2026, requiring people to be informed when they interact with an AI system. AI-generated or manipulated text, images, audio and video must carry machine-readable markings. Deepfakes, emotion recognition, biometric categorisation and AI-generated public-interest texts carry specific information duties.

Commission Implementing Regulation (EU) 2026/1755, published on 21 July 2026, establishes procedures for assessing general-purpose AI models and enforcing provider obligations. The Commission can request access to APIs, source code, model weights or hosting infrastructure and appoint independent experts. The rules also cover fines, urgent interim measures and providers’ rights to be heard and access their files.

Penalties for prohibited practices can reach 35 million euros or 7 percent of global turnover. Compliance duties extend beyond model providers to businesses deploying their systems, including banks, insurers and law firms. Article 26 establishes deployer obligations, while Article 4 requires organizations to train staff in AI literacy. Certain uses may also require fundamental-rights impact assessments.

Companies should inventory their AI systems, classify risks, update usage policies and review supplier contracts for conformity declarations and training-data summaries. Regulatory sandboxes are mandated across member states by 2 August 2027, offering start-ups and small firms fee-free testing. The deferred deadline for high-risk Annex III uses, including recruitment and credit scoring, is 2 December 2027.