This Week in AI: The Rules Finally Have Dates

The EU AI Act's Article 50 transparency obligations have been in force since August 2, 2026, which means customer-facing chatbots and synthetic media now carry real disclosure duties. The machine-readable watermarking requirement for generative systems already on the market kicks in December 2, 2026, and high-risk obligations for stand-alone systems follow in December 2027.

Italy introduced Article 437-bis into its criminal code, effective September 30, 2026, establishing criminal liability for failing to adopt required safety measures in high-risk AI systems. It's one of the first times an AI failure became a personal legal risk rather than a fine a company absorbs.

In the US, the FTC is weighing whether to hold platforms accountable for ad-optimization practices that amplify impersonation scams, and states kept moving: Connecticut added generative-AI subscription disclosure and algorithmic-pricing restrictions phasing in through 2028, while California's SB 947 and Colorado's AI Act remain live.

For builders, the practical takeaway is to map each system by role: provider, deployer, or both. Buying a compliant product does not make its use compliant, and the deployer duties, especially in hiring and credit scoring, land on the customer, not the vendor.

The 'the AI Act was delayed' reading is wrong in an expensive way. Only some deadlines moved. Transparency is already law, and ignoring it is the mistake that's cheapest to avoid and most expensive to make.