Article 50 Takes Effect August 2. Here's the Two-Week Readiness Checklist.

Most of the EU AI Act coverage this spring was about delay. The Digital Omnibus pushed high-risk obligations to December 2027 and August 2028, and plenty of teams filed the whole regulation under "next year's problem."

Article 50 is not next year's problem. The transparency obligations were never postponed. They take effect August 2, 2026, and if your product serves EU users, they very likely apply to you, because Article 50 covers the AI features almost everyone ships: chat interfaces, generated content, and synthetic media.

Here is what actually lands on August 2, what got a narrow grace period, and a checklist you can complete in two weeks.

What Article 50 requires from August 2

Four obligations, in plain terms:

1. Tell people when they're talking to AI. Any AI system intended to interact directly with people must be designed so users know they're interacting with AI, unless it's obvious from context. If your product has a chatbot, an AI assistant, or an AI voice on the phone, this is you.
2. Label AI-generated and manipulated content. Synthetic audio, images, video, and text need to be identifiable as AI-generated, including deepfake disclosure obligations for deployers.
3. Mark generative outputs in machine-readable form. Providers of generative systems must ensure outputs carry machine-readable marking (watermarking). This is the one place the Digital Omnibus granted slack: systems already on the market before August 2, 2026 have until December 2, 2026 to meet the machine-readable marking requirement, and only that requirement. New systems get no grace.
4. Disclose emotion recognition and biometric categorization. If you run either on people, they must be informed.

Penalties for transparency violations run up to €15M or 3% of worldwide annual turnover, whichever is higher.

What the Digital Omnibus actually changed

Worth being precise, because we've seen the timelines conflated in vendor marketing and even some legal summaries:

• Delayed: high-risk obligations for Annex III standalone systems (now December 2, 2027) and AI embedded in regulated products (August 2, 2028). Regulatory sandboxes move to August 2027.
• Not delayed: Article 50 transparency (August 2, 2026), the prohibitions and AI-literacy duties live since February 2025, and the GPAI obligations live since August 2025.
• Added: a new Article 5 prohibition on AI used to generate non-consensual sexual content and CSAM, with technical safeguards required by December 2, 2026. If you run generative models, that date now carries two obligations: watermarking grace expiry and safeguard deadlines.

If your compliance program paused when the delay headlines hit, the practical effect is that you now have two weeks of Article 50 runway instead of six months.

The two-week checklist

Days 1–3: Inventory your user-facing AI touchpoints. Not your model inventory, your disclosure surface. Every chat widget, AI email draft feature, generated image or report, voice agent, and recommendation explanation that reaches an EU user. Most teams find touchpoints in products that predate their AI program.

Days 3–5: Classify each touchpoint against the four obligations. Interaction disclosure, content labeling, machine-readable marking, emotion/biometric disclosure. Note which generative systems were on the market before August 2. Those get the watermarking grace; everything else doesn't.

Days 5–8: Close the disclosure gaps. Interaction disclosures are usually copy and UI changes. Content labels depend on where output leaves your system: email, documents, chat, API responses. This is where most of the two weeks goes.

Days 8–11: Wire the marking plan. For generative outputs, adopt provenance marking (C2PA or equivalent) or your model provider's watermarking where available, and document the choice. If you're on the December 2 grace, write down why you qualify. You'll want that documented.

Days 11–14: Start logging evidence. Article 50 compliance is a continuous property, not a launch-day property. A disclosure that a later deploy silently removes is a violation with no witness. Log every AI interaction disclosure and every content-labeling decision somewhere an auditor can replay.

Where enforcement fits

That last item is the difference between compliance on August 2 and compliance in March when someone asks you to prove it. A policy document says your chatbot discloses; an enforcement log shows every session where it did, and catches the deploy where it stopped.

That's the layer Aguardic covers: the EU AI Act policy pack evaluates AI outputs and integrations against Article 50's disclosure and labeling requirements in real time, blocks or flags violations per your enforcement mode, and writes the audit trail continuously. When the high-risk obligations arrive in December 2027, the same evidence architecture extends to Articles 9–15, which is also the honest way to use the delay: build the logging discipline now on the obligation that's live.

You can check your exposure in about two minutes with our EU AI Act audit. It classifies your systems against the current (post-Omnibus) structure of the regulation and tells you which Article 50 obligations apply.

Aguardic is not a law firm and does not provide legal advice. Verify obligations for your specific systems with counsel.