AI news
What's happening in AI
The AI news worth knowing, in plain English. New models, big moves, safety, and what it all means for business.
Summarised by AI, read by Andrew. Each night AI picks the most interesting AI stories and summarises them from the original reporting. I read every edition and correct anything that's wrong, and every story links to its source. How I use AI
Saturday 17 October 2026
7 stories
Applications
OpenAI has upgraded its ChatGPT Chrome extension so it can read your open browser tabs and, in Work mode, act on them. Users tag Chrome in ChatGPT to pull information from tabs, and OpenAI's demo compares sofas across three open pages. Work mode can send messages, fill out forms, download files and group tabs, while the sidebar keeps your full chat history beside the page you're viewing. The Verge notes that OpenAI's ChatGPT Atlas browser is gone, and neither source mentions Australian availability.
Why it matters: If staff compare suppliers or fill in online forms, the extension works on all ChatGPT plans. Set rules before Work mode starts sending messages on their behalf.
Source: The Verge · Also reported by The Decoder
Safety & policy
Meta's updated commercial terms, effective 30 October, appear to let it crawl websites linked to business accounts and use that content to train its AI products, according to Nieman Lab. Legal experts quoted by Nieman Lab say the terms could be used to argue they override robots.txt blocking instructions, a reading Meta disputes. A Meta spokesperson says there is no change to how it uses web crawlers, and that site owners can still use robots.txt. Businesses that refuse the terms may lose access to Meta's products for commercial purposes.
Why it matters: If you run a business page on Facebook or Instagram, check whether your website could be crawled under these terms. Review your robots.txt settings before 30 October.
Source: The Verge · Also reported by Ars Technica
Safety & policy
The US Federal Trade Commission has subpoenaed OpenAI and Anthropic over recent safety incidents with their AI models, a spokesperson confirmed to CBS News. Axios reports the demands seek documents on security incidents, safeguards and privacy-related marketing. An Axios source said the agency plans to seek testimony from top executives and is exploring possible consumer protection violations. Eight state attorneys general sued OpenAI this week on consumer protection grounds. Neither company responded to Ars Technica's request for comment.
Why it matters: If your team uses ChatGPT or Claude, check how your provider reports security incidents and what its privacy terms promise, as regulators are now asking these questions.
Source: Engadget · Also reported by Ars Technica
Safety & policy
Australia's Digital Duty of Care bill would force social media platforms to let users turn off recommended content, so they see only accounts they follow. Announced in September, the algorithm switch stops short of defaulting to off, as campaigner Chanel Contos had sought. Breaches could attract fines of up to A$100m, but the Trump administration and some Australian right-wing parties say the reforms threaten free speech. Meta, Snap, TikTok, X and Google have declined BBC requests for comment.
Why it matters: If you post or advertise on social media, fewer non-followers may see your content if users switch off recommendations. The rules aren't final, so watch how they settle.
Source: BBC Technology
Gaming & entertainment
Sega and Creative Assembly have temporarily withdrawn Total War: Shogun 2 – Complete Edition from sale, within days of its 13 October launch. Steam reviews are overwhelmingly negative, with players pointing to AI upscaling and artwork showing flaws such as extra fingers. In a joint statement, the companies admit the game falls short of their standards and say they will reinstate the original game while they investigate. Existing owners can still access and play the remaster.
Why it matters: If you publish products with AI-generated visuals, check quality before launch. Customers spot flaws quickly, and a public pull can damage trust and sales.
Source: GamesIndustry.biz · Also reported by The Verge
Safety & policy
Advertisers can target ChatGPT users by the topics they ask about, or by uploaded lists of emails, phone numbers or Android advertising IDs. Reporter Joseph Cox ran tests for 404 Media, which bought two ads inside the chatbot. OpenAI says topic keywords only guide matching and aren't exact-match rules, and that it deletes uploaded files, usually within 24 hours. Lists above 25,000 entries can target people, while smaller lists can only exclude them, and 404 Media says turning off ads personalisation doesn't stop ads entirely.
Why it matters: If your team uses ChatGPT for work, the ads personalisation switch doesn't fully stop tracking, so keep sensitive client details out of chats.
Source: The Neuron
Safety & policy
China's commerce and science ministries have released a draft update to its export control list, covering 23 key AI technologies. Large models face controls at 10^26 FLOPs of training compute, 2 trillion parameters or a global top-10 ranking. Tools that autonomously write or alter malware to attack critical infrastructure face a total export ban. Publishing restricted technology as open source qualifies for automatic licensing, and companies must register exports through a commerce ministry portal.
Why it matters: If you use Chinese AI models through apps, APIs or SaaS tools, the draft leaves that use untouched. It isn't yet clear how it treats foreign businesses that install the largest models on their own servers.
Source: The Decoder