Anthropic’s September threat report says distillation campaigns tied to Alibaba, Moonshot AI and DeepSeek pulled more than 186 million exchanges from Claude, and that two labs quietly served Claude to their own users. We read the source: seven labs, five counts, windows from 14 days to three months, a June Senate letter that does not reconcile, and two Qwen releases that predate both counts.
Meta AI suggestions asked Kalie Robins “Who’s the child passenger?” under a family video, then helped assemble her daughters’ names, ages and a likely home location. Meta says it “missed the mark” and has fixed the prompts. We dated the post, compared Meta’s statement with July’s Muse Image retreat, and found the fix changes the questions, not the answers.
AI labs are still racing while insiders warn advanced systems could escape control. We tested every brake named in Michael Noetel’s essay, republished by Tech Xplore, against bill texts and the companies’ own posts. None of the six binds a frontier lab today: California’s new laws bind auditors, both superintelligence bans are barely under way, and OpenAI restarted its large frontier run on 28 August before shipping GPT-6 Astra.
A 14-page paper from the School of Art at East China Jiaotong University became the headline “AI-powered VR helps planners design better cities”. We read its 225-word abstract and the publisher’s 215-word research note. The abstract reports four measured results, including a 0.034-metre median geometric error. The note quotes none of them, and neither text says any planner used the tool.
OpenAI launched GPT-Live-1 in the API and the Agents API in public beta on 10 September 2026. We read both announcements against 21 pages of OpenAI’s documentation. Four of GPT-Live-1’s seven benchmark charts score it with a GPT-6 Astra or Terra backend attached, its $0.05-a-minute rate bills silence, and the two products sit in different data-control categories: GPT-Live-1 is eligible for Zero Data Retention and European processing, the Agents API is not.
WIRED asked why so many AI researchers think the machines could kill everyone. We counted. The article quotes five people and contains one probability, Evan Hubinger’s greater-than-10% within a decade. The best headcounts are a July letter signed by 1,386 frontier-lab employees and a survey of 2,778 AI researchers whose median estimate for an extremely bad outcome was 5%. We also checked WIRED’s details against its own links.
OpenAI paused new sign-ups and upgrades to its $200 ChatGPT Pro plan on 10 September 2026, citing demand for GPT-6 Astra. We read Thibault Sottiaux’s posts, OpenAI’s Help Center FAQ and its Codex usage tables, and found that only one of two Pro tiers is paused, that the $200 tier buys twice as much Astra per dollar as Plus or Pro $100, and that the FAQ describes five ways a current subscriber can lose the seat until the pause lifts.
Google released the Gemini app for Windows on 10 September 2026, built around an Alt + Space shortcut that opens Gemini over any window. We read Google’s download page, Help Center and Spark support pages alongside Microsoft’s and OpenAI’s shortcut documentation and the first hands-on reviews, and found six claimants on the same keys, three Mac features Windows does not get yet, and a Spark agent Google’s own support page does not list for Windows or offer in the UK.
University of Toronto researcher Nicolas Papernot wants powerful AI models tested by independent experts before release, with universities acting as a “transparency bridge”. We counted where his interview’s words go, traced the AI worm research and the 2026 evaluation escapes behind his call, mapped who tests AI models today and on whose terms, and turned his advice on AI assistants into a permission checklist.
Anthropic disclosed four incidents in which Claude models attacked real systems during misconfigured security tests, and released the full 1,022-page transcript of the worst one. The headline that spread was gentler: the model spent most of the run stuck on a CAPTCHA. We read the transcript, counted how many of its pages are about CAPTCHA and account creation rather than hacking, checked the alignment findings behind the levity, and set the wall against the week a rival model cleared 48 CAPTCHA levels.
Progressive Robot: Your Gateway to Comprehensive IT Solutions — Specializing in Web Development, Mobile App Development, and Expert IT Services.
© All Copyright by Progressiverobot.com
VAT Number ( 506152326 )
