Launch a high-risk AI system before Dec 2027 and it's grandfathered from the EU AI Act's high-risk obligations.
However, one "significant change" to the design and the grace is gone.
The bans and transparency rules apply either way.
https://t.co/CbNrctveUf
When your time comes to register under Article 49, make your submission accurate, but write it knowing it can be read by your competitors, journalists, marketers... anybody.
https://t.co/dHmCHOUHYi #EU_AI_Act
EU AI Act Article 49 requires Annex III high-risk systems be registered in a central EU database. Most of the information will be public. It will effectively be a live index of high-risk systems placed on the market, visible to people who are NOT regulators.
Anthropic now wants a selfie and government ID.
Is this the price of admission to frontier AI?
Using the most capable software increasingly means proving who you are first.
Two different things in the EU AI Act are both called "transparency". Only Article 13 transparency was delayed.
Art 50 transparency relates to chatbot disclosure and content marking. Applies from 2 Aug 2026.
Art 13 transparency of high-risk systems - from 2 Dec 2027.
The Digital Omnibus didn't delay AI transparency.
It moved the high-risk obligations to Dec 2027 and cut the grace period for marking of AI-generated content to 2 December 2026.
If you run customer-facing chatbots, your first deadline is still 2 August 2026.
Seeing a lot of "the EU AI Act is delayed to 2027, relax" posts. That's half true. Article 50, AI disclosure, was not delayed and it applies to most customer facing AI systems. It applies from 2 August 2026.
That's just 7 weeks away.
The EU Commission recently published draft guidelines on transparency requirements under the #EU_AI_Act. There's 2 obligations here. You have to mark AI generated content AND make it machine detectable. There's some engineering required. Read more here⬇️⬇️⬇️
Article 50 transparency looks easy until you realise it applies to all AI outputs in any format. JSON API responses, RSS feeds, audio, image, video. The carrier doesn't change the obligation.
https://t.co/hzoAfnwQJt #AIAct
If you run a high-risk AI system, will you be able to provide a "clear and meaningful explanation" to a user on how a decision was reached? It's a requirement under the EU AI Act. “The AI said no” is not an explanation. https://t.co/IPJnzSgL6o
Time and again we see organisations wrongly assume they are a deployer and not a provider under the #EU_AI_Act. If you add a UI or wrap an AI model into your high-risk product, you are the Provider of the system and must address your obligations.
GPT-4 is not an AI system under the EU AI Act. Neither is Claude. They're AI models.The distinction matters.
Compliance obligations apply to the company that built the system, not the company that trained the model.
BREAKING: It's official. The EU AI Act enforcement deadline has been delayed to 2 December 2027, giving providers of high-risk AI systems an additional 18 months to meet their compliance obligations. #AIact#AI_Act
EU AI Act Article 11 Technical Documentation pulls every other Section 2 obligation into a single producible record. Get started now on your obligations. https://t.co/G6r86sZhli #AIact
EU AI Act Article 11 is the most important doc required by the Act. It must “demonstrate that the high-risk AI system complies with the requirements” and provide “all the information necessary to assess the compliance of the system.”
Deadline is 2 August, 2026.
EU AI Act Article 11 Technical Documentation pulls every other Section 2 obligation into a single producible record. Get started now on your obligations. https://t.co/G6r86sZhli #AIact