This week's AI news that actually matters to a small or mid-sized Estonian company — practical angles, no hype.
OpenAI shipped a mid-tier model built around screen work and reading visual material
The emphasis is on handling documents, drawings and forms plus operating software, rather than chat alone. Why it matters: that is exactly the work — writing proposals, filling forms, interpreting drawings — that eats hours in small teams, and the price tier fits an average budget. simonwillison.net
Moonshot released Kimi K3.1 Pro, promising near-frontier quality at roughly half the cost
Open weights are promised for early November, which also opens the self-hosting route. Why it matters: competition is pushing prices down — worth re-checking your AI bill and testing whether the same output costs half as much elsewhere. simonwillison.net
Google's revised privacy terms allow Gmail content to feed advertising targeting
It applies in particular once the Personal Intelligence features are switched on. Why it matters: if company correspondence lives in Google accounts, review the settings and your legal basis — this is a GDPR and client-contract question. arstechnica.com
Apple's latest OS update adds Siri features and a way to restrict which apps the assistant can reach
Access can now be granted or denied app by app. Why it matters: with business data on staff phones, this gives you a concrete lever for setting internal AI usage rules. arstechnica.com
Anthropic filed to go public, devoting a large share of the document to risks
The prospectus is a rare look at how the company itself assesses its technology and its downsides. Why it matters: if you depend on a vendor for critical processes, know its funding and risk profile — public-market pressure usually reshapes pricing too. technologyreview.com
Chatbots are sending stuck customers to the wrong place because real support is unreachable
One publication received a wave of help requests from people whom ChatGPT had pointed their way. Why it matters: a warning for anyone planning to hand all support to a bot — wrong routing produces frustrated customers and reputational damage. simonwillison.net
Analysis: AI policy fixates on catastrophes while everyday harms go unaddressed
Companies can clear low minimum bars and still ship products that hurt users. Why it matters: accountability lands on whoever deploys the tool — measure real customer impact, not just paper compliance. technologyreview.com