Uber fined for automated driver deactivations, BBVA fined over one ignored marketing objection, Anthropic pauses customers in error, a surprise Claude bill and
By Avik Ghosh, Managing Founder, illuminis
Five stories from the past few weeks that matter to anyone buying, running or regulating AI, with our take on each.
On October 8, the European Data Protection Board published the decision of the Dutch data protection authority, the Autoriteit Persoonsgegevens, to fine Uber €824,990,000. The authority found that Uber made fully automated decisions about drivers, temporarily deactivating accounts on suspicion of fraud or low customer reviews and permanently deactivating them for persistently low reviews, which cut off their income through Uber. It also found that Uber did not tell drivers enough about the automated decision-making. The case began with complaints from 171 French drivers and covers incidents between 2018 and 2022. Uber has stopped the practices and has appealed the fine. Source.
Our take: This fine was imposed under data protection law, before any AI Act penalty. Regulators are already asking which decisions a system made on its own, what people were told, and where a human stepped in. Any company using automated systems to decide who gets work, credit or care needs to answer those questions with records that cannot be quietly changed afterwards. See InferTrust™.
On October 9, the European Data Protection Board published an Italian data protection authority decision, dated July 3, fining the Italian branch of BBVA €5,508,000. A customer objected to direct marketing in the bank's app and again through Customer Service, but kept receiving promotional notifications from October 2025 to May 2026, at least ten in total. BBVA said the choice was recorded but a technical failure stopped it syncing with the CRM unit that sends commercial messages. The authority rejected BBVA's argument that the customer should have used email addresses listed in its privacy policy, and ordered the bank to report its corrective measures within 30 days. Source.
Our take: The case involved one person and one sync failure, and the fine still ran into millions. An objection recorded in one system but not honoured in the system that sends the messages counts as not honoured at all. Marketing teams need contact data whose permissions travel with it. That is why RelationLens shares a contact only when the employee approves it, and forgets what nobody reviews.
On October 7, Anthropic reported that an issue caused some organizations to be incorrectly paused for reaching their spend limit. Their requests were refused across the Claude API, Claude.ai, Claude Code, Claude Security and Claude Cowork. Anthropic marked the issue resolved at 21:23 UTC that day. Source.
Our take: Not every outage is a server problem. A billing control at the provider failing can stop work just as completely as a crashed model, and the customer can do nothing about it. Companies and software vendors that depend on a single provider absorb every one of these events. Model-level failover belongs in the plan, and in every SaaS vendor's sales story. See the Embedded License.
On October 9, The Register reported that Norwegian startup Vegalabs received $25,060 in Azure credits through Microsoft for Startups and deployed Claude in Microsoft Foundry in August, expecting the credits to cover it. Microsoft's sponsorship rules exclude Anthropic models bought through Azure Marketplace, and the invoice reached $17,600 before tax while the sponsorship portal still showed credits available. $21,168 in unused credits expired on September 8. Vegalabs acknowledges it did not read the exclusions or set a budget alert, and says Microsoft and Anthropic support each sent it to the other for a waiver. Source.
Our take: The same model can be billed differently depending on where you buy it, and the dashboard you watch may not show the charge that matters. Finance teams should know what each model costs, through which channel, before the invoice arrives, not after. Cost to finish the work is the number to manage. That is the case for CompletionPrism™.
On October 8, the U.S. Court of International Trade certified a class of importers who paid tariffs under the International Emergency Economic Powers Act between February 1, 2025 and February 19, 2026, excluding those with their own lawsuits. The court has ordered refunds of the roughly $166 billion collected, and Customs and Border Protection has paid out over $100 billion through its CAPE portal. The court said about $30 billion would otherwise stay with the government, much of it owed to small importers who have not used CAPE. A closed conference is set for October 13 and proposed scheduling is due October 22, while a Justice Department appeal on refunds for finally liquidated entries is still pending. Source.
Our take: Refunds go to importers who can show what they paid, on which entries and when. Procurement and finance teams that cannot connect duties paid to suppliers, shipments and contracts will be slower to claim money they are owed, and slower to reprice when the rules change again. Sourcing decisions should rest on true landed cost. See SourceMind™.
In the News comes out every week. Subscribe from the Insights page to get it by email.