WhatsApp had hidden 102 instruction blocks for its support chatbot across 80 help pages, one of which told it not to suggest contacting a data protection authority. Meta calls it a bug, an explanation Swiss company Proton does not accept.
In a Guardian opinion piece, Chris Stokel-Walker notes that services in the Netherlands, Finland and at Wikimedia were switched off deliberately because staying online was too risky. He argues that a few overstretched developers cannot secure the web alone and that the companies profiting from it should help pay.
In a Verge interview, Microsoft cofounder Bill Gates discusses the AI era he described in an essay published in late August. He warns of two decades of turbulent disruption.
AI has become a major issue ahead of the November 3 US midterms, according to Next. AI data centers in particular have turned into an electoral liability for Republicans.
The authors of the Laws of Armed Conflict want to set guardrails on AI weapons systems. 404 Media asks whether governments will heed them.
The GSA, the US federal purchasing agency, was due by October 19 to add a clause to new tenders for LLMs, chatbots and agents: no training on government data, deletion of fine-tuned weights and embeddings at contract end, and incident alerts within 72 hours. European vendors can still bid but must give 30 days' notice of any adaptation to non-US rules that substantially changes their model's answers, which could cover AI Act tuning.