Commission adopts Article 50 transparency guidelines — 12 days before the rules bite
The European Commission adopted its guidelines on the transparency obligations for providers and deployers of AI systems on 20 July 2026 — the first Commission instrument interpreting the full scope of Article 50, covering chatbot disclosure, deepfake and synthetic-media labelling, and AI-generated text on matters of public interest. They are non-binding, but they define the scope competent authorities will read Article 50 against once the obligations apply on 2 August, and they land alongside the separately-assessed voluntary Code of Practice on Transparency of AI-Generated Content (formally judged 'adequate' by the Commission this month). The practical read for anyone running a chatbot or shipping synthetic image, audio, video or text into the EU: the interpretive gaps you have been guessing at — what counts as a 'directly interactive' system, which disclosures sit with the provider versus the deployer, and the artistic and law-enforcement exceptions — now have an official answer, and you have under two weeks to reconcile your notices and content-marking to it before €15M / 3%-of-turnover exposure switches on. Note this is a different instrument from the still-pending Commission/EDPB guidance on the GDPR × AI Act interplay — do not conflate them.
Source: European Commission — Shaping Europe's Digital Future
Judge grants final approval of Anthropic's $1.5B authors copyright settlement
US District Judge Araceli Martínez-Olguín (San Francisco) granted final approval on 20 July to Anthropic's $1.5 billion settlement with authors whose books were used to train Claude — the largest copyright settlement in US history, at roughly $3,000 per infringed work across about 500,000 titles, with objections that the sum was too low overruled. The important distinction for advisers is what this does and does not decide: it resolves the piracy/acquisition claims — that Anthropic sourced pirated copies — not the fair-use question on training itself, which an earlier ruling had treated as transformative where the underlying books were lawfully acquired. So the case does not bless or bar AI training on copyrighted works; it prices the how-you-got-the-data question, and prices it enormously. The reusable lesson for any client building or fine-tuning on a corpus is that provenance and the acquisition trail are now the litigable centre of gravity — the same point the EDPB's 8 July web-scraping guidelines make on the data-protection side. Expect this figure to become the anchor other rightsholder classes negotiate against.
Source: TechCrunch
Unsure which of these developments applies to your AI systems?
Begin in writing →This briefing is general information, not legal advice, and does not create an advisor–client relationship. Summaries are original; follow source links for the full record. Adesanya AI Advisory — Abdulwahab B. Adesanya, LL.M in Intellectual Property & IT Law (UCD).