Enterprise AI Radar — Tuesday, July 21, 2026
noticias radar ia claude ai act

Enterprise AI Radar — Tuesday, July 21, 2026

· CompaniesAutomation

Anthropic ends the Fable 5 limbo: it stays in Max and moves to credits in Pro with $100 included. Brussels publishes the AI Act transparency guidelines two weeks before obligations start, and SAP closes the acquisition of Prior Labs for over $1 billion to bring AI to tabular data. Practical reading for your company in three news items.

The Sunday radar wondered if Anthropic would extend Fable 5 for a fourth time; it didn't, and the limbo ended yesterday with a split into two: those who pay for Max keep it, while those on Pro move to credits. Meanwhile, Brussels published the final transparency guidelines for the AI Act less than two weeks before the obligations go into effect, and SAP closed the acquisition of Prior Labs to build a frontier AI laboratory for tabular data in Europe. Price, compliance, and business data: the three fronts of the day.

Fable 5 exits limbo: stays in Max and moves to credits in Pro

Anthropic ended the extension soap opera on Friday and the distribution was executed yesterday, Monday: Max and Team Premium plans keep Claude Fable 5 as a permanent part of the subscription — with half of its weekly limits —, while Pro and Team Standard receive a one-time $100 credit and then pay per use, at $10 per million input tokens and $50 per million output tokens, with a 90% discount on cached input. After three deadlines (July 7, 12, and 19) and two last-minute extensions, Anthropic's most capable model is out of the $20 plan. For your company: if you have workflows running on Fable 5 in a Pro plan, the $100 credit is your testing ground — measure what each real task consumes before burning through it and decide with numbers if it's worth jumping to Max, prepaying credits, or locking in Opus 4.8 — and watch out for any production agents changing models without your knowledge. Source

Brussels publishes AI Act transparency guidelines: your chatbot has two weeks to identify itself

The European Commission published the final guidelines yesterday on the transparency obligations of Article 50 of the AI Act: chatbots must identify themselves as machines to the user, AI-generated content must be marked, and deepfakes or AI-generated texts of public interest must be labeled. These obligations apply from August 2 and, unlike the high-risk ones — postponed to December 2027 by the Digital Omnibus —, they have not moved. For your company: there are less than two weeks left — if you have a customer service chatbot, add the notice that they are talking to an AI now; if you publish generated content (images, audio, video), prepare the marking. The guidelines are not binding, but they are the criteria that the inspector will use to interpret the regulation. Source

SAP acquires Prior Labs: the AI coming for your ERP tables, not for the chat

SAP closed the acquisition of Prior Labs on Friday, the German startup specialized in tabular foundation models, 18 months after its founding, and has committed to investing over 1 billion euros over four years to turn it into the European frontier lab for AI for structured data; Prior Labs will continue to operate as an independent entity. Its TabPFN model, published in Nature and with more than three million downloads, makes predictions on tables without task-specific training. For your company: the signal is that the next wave of enterprise AI is going after the data you already have in the ERP, CRM, and spreadsheets, not the chat; TabPFN is open source and you can try it today for predictions on your tables — defaults, demand, customer churn — without training any models or moving data out of Europe. Source

What to watch tomorrow?

Open weights week: DeepSeek promises the stable version of V4 for Thursday 24, and Moonshot the Kimi K3 weights for the 27th; also pending to see if Google covers the third delay of Gemini 3.5 Pro with a bridging 3.6 Flash.

Watch the 1-minute video