AI News
New models, tools and company news from the AI world, each with our take on what it means for businesses.
- Instagram Cracks Down on Undisclosed AI ProfilesInstagram will restrict the reach of AI-generated profiles that skip its new mandatory label. Fake doctors and crypto-promoting AI influencers are the target; everyday AI tools are not.
- NPR Tested 6 Chatbots on Propaganda; Most Held the LineNPR and NewsGuard tested six chatbots against 30 questions built from Russian, Chinese and Iranian false narratives. Bots scored 75%; search summaries did worse.
- Glassdoor: Employee AI Sentiment Fell From 81% to 43%Glassdoor scanned seven years of employee reviews mentioning AI: positive sentiment fell from 81% to 43%. Who soured first, and how solid the numbers are.
- AWS Orders 2 Million More Nvidia GPUs for the Agent EraAWS and Nvidia will deploy 2 million additional GPUs in 2027-2028, on top of existing plans, including a 100,000-GPU AI factory for the US government.
- OpenAI Is Buying Tens of Thousands of Mac Minis for AgentsThe Information reports OpenAI bought tens of thousands of Mac minis to train computer-use agents with reinforcement learning, while Anthropic rents Macs via AWS.
- Infostealer Malware Is Hijacking Paid Claude SessionsAnthropic warns that Vidar, LummaC2 and other infostealers are stealing Claude session cookies, bypassing passwords, 2FA and SSO. The cleanup order matters.
- Anthropic Trims Claude Code Limits by 17% in PracticeA permanent 25% increase to Claude Code's weekly limits replaces a temporary 50% boost on September 14. Net effect for users: 17% less capacity than last week.
- Survey: 3% of US Workers Lost a Job to AI; Call Centers HireA YouGov survey of 1,250 employed Americans finds 3% lost a job to AI since 2023 and 6% found one because of it. The same day, Apollo's chief economist pointed out that Philippine call-center employment has doubled to 2 million.
- DeepMind Runs Double-Blind Evals, Co-Scientist Runs the LabIn one week Google DeepMind ran the first double-blind evaluation of a frontier model, where neither weights nor test prompts are exposed, and published a Co-Scientist that controls lab equipment and grew semiconductor films on the first try.
- Turkey Plans AI Factories and Growth Zones With 25B LiraTurkey's 2026-2030 AI Action Plan now has an implementation layer: pre-permitted AI growth zones, shared-compute AI factories, two funds totalling 25 billion lira and 1,000 AI vouchers for small businesses. The first zone is due within a year.
- Researcher Bypasses Claude Code Auto Mode 80% of the TimeJohann Rehberger chained a website summary request, a zip archive and Python module shadowing to get code execution through Claude Code's Auto Mode in 80% of runs. In some runs the safety layer then blocked Claude's own cleanup.
- Anthropic's Model Hardware Standard Puts Agents on Lab GearAnthropic opened a research preview of the Model Hardware Standard, a shared spec for AI agents to operate microscopes, liquid handlers and robot arms. Partners report integrations dropping from weeks to hours, and one telling failure.