When AI Disclosure Is Due

The European Commission published guidelines on 20 July 2026 to help providers and deployers meet the AI Act's transparency obligations. Those obligations apply from 2 August 2026. The stated aim is that people can recognise AI interaction and AI-generated or altered content, so fluency stops masking deception. The operator problem is narrower: inventory the live systems that talk to customers or the public, and score each one for notice, mark, editorial ownership, and a stop rule before the August date.

Article 50 transparency is a workflow tax. Budget the process now; an incident later is the unplanned bill.

What Changed This Week

Article 50 of the EU Artificial Intelligence Act covers transparency for certain interactive AI systems and for AI-generated or manipulated content. The Commission's July 2026 transparency guidelines spell out provider and deployer duties. Providers must design interactive systems so users know they face AI, and must support machine-readable marks that help detect generated or manipulated content. Deployers must inform people about deepfakes; about AI-generated public-interest content that lacks human review or editorial control; and about emotion recognition or biometric categorisation systems.

Calendar beats vocabulary. 2 August 2026 is close. Other AI Act chapters have been debated and delayed, yet public reporting has kept Article 50 on the August clock. Operators who banked on a general "regulation got postponed" story still face a customer-facing disclosure duty on that date.

U.S. and Asian companies still meet this when the system is offered in the EU, when EU users are in scope, or when enterprise buyers demand EU-grade disclosure as a commercial term. The same screen becomes a sales and product job even when the legal home office sits elsewhere.

What this changes for operators

Three operating ideas matter more once the calendar is fixed: write the rules before you pick tools; treat disclosure as a workflow, not a footer; and name who owns residual risk when public speech goes wrong. The July guidelines put those ideas on a hard date.

Rules before tools. Teams that only compare models this quarter will bolt a disclaimer onto a product already in market. A one-page set of rules should already list AI-assisted customer conversations, public statements that need human editorial control, and content classes that carry a mark before they leave the firm. See also Decide the Rules Before Buying Tools.

Disclosure as workflow. When to Disclose AI Use tracks generation, travel path, approval, and the point where missing disclosure must stop publication. Article 50 of the EU AI Act (see the Commission's transparency guidelines) makes the first half of that checklist mandatory for in-scope systems. The mark and the user notice are the public edge of generation, review, routing, and publication.

Residual ownership. When a chatbot uses the brand voice and a customer claims deception, cost settles on the deployer who put the system in front of customers, not cleanly on "the model vendor." Transparency lowers some of that risk, but a named owner of the workflow remains required. See Where AI Coverage Still Fails.

Four Operator Mistakes

Mistake one: marketing-sentence disclosure. "Powered by AI" in a footer fails interactive systems. Users need to know at first contact that the counterpart is a machine. Product, support, and legal must agree where the notice sits and which wording stays honest.

Mistake two: treating public-interest content like ordinary drafts. Guidelines and surrounding commentary weight AI-generated public-interest material that lacks human review or editorial control. A research blog, market note, or founder essay that ships heavily machine-written work without a human owner sits in a different risk class from an internal brainstorm. Editorial control is a named control.

Mistake three: leaving agents out of scope. Agentic systems whose outputs users perceive directly fall under the same transparency logic. An agent that emails customers, posts content, or speaks through a channel that looks human needs disclosure as part of the action rule, written before launch.

Mistake four: waiting for perfect insurance or a perfect policy product. The argument in Where AI Coverage Still Fails still applies. Regulatory disclosure and insurance residual risk are separate problems. Hoping the market invents a complete product before August solves neither.

A One-Week Operator Screen

Before 2 August, score every customer-facing or public-facing AI surface on the five rows below.

What This Digest Is For

Earlier memos cover rules before tools, reality tests for AI strategy, disclosure checklists, accountability gaps, and insurance coverage gaps. This digest runs one market event through that frame so an operator can decide this week. Start with When to Disclose AI Use and Is This AI Strategy Real? if you need the longer checklists.

Inventory the five highest-blast-radius AI surfaces that talk to customers or the public. For each surface, name a disclosure owner, a notice design, a content-mark rule when media is involved, and a stop rule when those pieces are missing. A blank inventory means the company is unready for a rule that treats silence as a liability.

Model releases will keep arriving. 2 August is a product deadline of another kind: whether the company can state that its machines are machines while they speak.