On 2 August 2026 the EU AI Act entered its enforcement phase, giving the European AI Office power to act against general-purpose AI model providers, a development that lands directly on the compliance desk of any Chinese model maker with global ambition.
Three tiers of pain
The penalty structure runs in three bands. Use of banned, unacceptable-risk AI can draw up to 35 million euros or 7 per cent of global annual turnover, whichever is higher. Most compliance and regulatory breaches, including those for general-purpose model providers under Article 101, reach 15 million euros or 3 per cent. False or misleading information to authorities caps at 7.5 million euros or 1 per cent. The decisive detail is that fines scale with global turnover, not EU revenue, so a company with small European sales is still exposed on its worldwide book.
Two regulatory maps
China’s route was service-first and label-first, with interim generative-AI measures from August 2023 and dedicated AI-content labelling rules from 2025. The EU built a risk-tiered law, banned through minimal, with enforcement concentrated in one body and pointed at the model layer. Chinese vendors already have content-labelling engineering in place, so the new work is model-layer systemic risk assessment, major incident reporting, cybersecurity and technical-document retention, not a start from zero.
The Brussels effect
The law’s out-of-region pull is the part that matters most. When a market is large enough, exporters adopt its standard directly, and bigger vendors tend to align one global product to the strictest rule rather than maintain parallel versions. That turns the EU gate into a global capability filter. For Chinese teams, the lesson after 2 August is that going abroad now includes a new line item: how fast you can make compliance a routine part of the build.
Read the original report (Sohu IT)
Translated and adapted from Sohu IT (it.sohu.com).