AI Content Labeling Is Now Legally Required on Both Sides of the Atlantic, but Less of It Than the Headlines Suggest
August 2, 2026 was billed for years as a hard enforcement cliff for the EU AI Act. What actually took effect that day is narrower than the advance coverage suggested, and knowing the difference matters if you're building or buying AI products that touch either the EU or California.
What actually became enforceable on August 2
The European Commission's AI Office began enforcing the AI Act's Article 50 transparency duties: AI systems must disclose when a user is interacting with AI, and AI-generated or altered content, including deepfakes, must be marked as such. Fines for violations reach up to €15 million or 3% of global annual turnover, whichever is higher. Anthropic responded by signing the EU's Article 50(2) Code of Practice, committing new Claude models released in the EU to support machine-readable content marking from launch.
What did not take effect: the bulk of the Act's high-risk system obligations. A Digital Omnibus approved by the Council on June 29, 2026 deferred those requirements by 16 months, to December 2, 2027, for standalone high-risk systems, and by 12 months, to August 2, 2028, for AI embedded in already-regulated products. If you were operating on the original compliance timeline for high-risk obligations specifically, that deadline moved. The transparency rules did not.
California moved the same week, for a narrower target
California's SB 942, the AI Transparency Act, took effect August 2, 2026 as well, after being delayed from its original January 2026 date. It requires large generative AI platforms to provide free AI-content detection tools and embed watermarks in generated media. Unlike the EU rule, this specifically targets large, consumer-facing platforms generating image, video, or audio content at public distribution scale, not general B2B SaaS tools broadly. Texas's Responsible AI Governance Act is being implemented in the same period with a meaningfully broader scope, applying to any developer or deployer doing business with Texas residents.
What to actually check
If you're building consumer-facing generative content features, both the EU marking requirement and California's watermarking rule likely apply to you now, not eventually. If your product is high-risk under the EU's Annex III classification specifically, and not a transparency-duty case, you have until December 2027, not August 2026, though product timelines rarely align neatly with regulatory ones, so treating that as a soft deadline rather than a distant one is the safer read.