The Daily Brief archive
Inference Wire publishes a five-minute AI intelligence brief every weekday morning. Read the latest edition, or work back through every edition we have published.
All editions
Newest first · page 2 of 5
AI is answering from years of contradictory reports, and $210 of agents opened five accounts
Assistants that finish tasks are arriving in Slack, Teams and Google's desktop app, and ChatGPT can turn a hand drawing into a finished image.
Claude can send email from a Gmail inbox, and Microsoft fixes a record 972 flaws
A judge has ruled Workday can be treated as an employer because its software does the screening, and Wispr Flow's new notetaker records meetings without joining them.
xAI's business agents sign into company software, and Kenya's essay industry is gone
An AI now produces the shortlist of options a leader reviews, and stolen tokens from one sales chatbot opened hundreds of Salesforce systems.
GPT-6 Astra halves ChatGPT message limits, and reviewing AI work becomes the expensive part
Hidden instructions in a document still capture OpenAI's newest model about one time in twelve, and Meta will cut Muse Spark prices about 95% for the right to train on customers' prompts.
Copilot users saved two hours a week on email, and Meta drops its AI usage score
Malware is stealing logged-in Claude sessions, and freelance listings for fixing AI work reached 10,760 in ten months.
AWS pushes software partners to charge per result, and a 45% price cut became a 20% bigger bill
e.l.f. Beauty's AI drafts up to 90% of its social media replies, and 40% of Kier's 11,000 construction staff open Copilot daily.
Only 7% of managers use AI in team meetings, and an AI break-in took ten hours
Google folds its Pics image editor into Workspace at no extra charge, and OpenClaw's new sharing feature keeps no one's access separate, by its own setup notes.
Shoppers sent by AI assistants buy 42% more often, and claims adjuster jobs fell 21%
OpenAI now bills some large customers only for finished work, and a Claude Code subscription buys about 17% less capacity from September 14.
Managers found 18% fewer errors when a named AI did the work, and Claude can now run lab robots
Gartner expects more than 40% of AI agent projects to be canceled by 2027, and Caterpillar is spending $100 million to retrain 118,000 workers.
The priciest AI setup performed worst on invoices, and nine in ten firms report no productivity gain
Claude's desktop app now signs into supplier portals through its own browser, and AI agent traffic to websites grew 7,851% in 2025.
Free AI dictation arrives in Chrome, and OpenAI's own agents went twelve days unnoticed
Meta calls off a plan to shrink teams by up to 60%, and Google adds a spending cap to Gemini Enterprise, where agent work is billed apart from the seat.
Chinese hacking groups doubled their attack volume with DeepSeek, and Grok's agent now costs $40
Forrester maps which technology markets AI shrinks first, and one job opening now draws 242 applications.
A $20 ChatGPT agent needs the whole inbox, and C.H. Robinson's staff is down a third
Travelers routes routine insurance questions to a model it built itself, and Alibaba's Wan3.0 turns a slide deck into a 30-second video for $3.
Only 26% of firms can say what AI costs, and the ones who can see returns five times as often
Replit's Free Mode runs everyday work on GPT-5.6 Luna at no extra charge, and only 1 in 5 companies say they are ready to redesign work around AI agents.
Everyday AI search tools push whole teams toward the same handful of ideas
Researchers took over five AI browser assistants with one email, and the best AI agent finished 26.2% of 1,490 real work assignments in a Berkeley test.
ChatGPT and Claude now learn a work task from one screen recording
Five agencies warn of AI-written attacks on Siemens plant controllers, three weeks after more than 30 Minnesota water systems were disrupted.
Only daily AI users feel good about their job security, and most employers have no AI rule
Microsoft Copilot revealed an undocumented parameter that ran prompts with no click, and Z.ai reports 2,436 flaws found across 269 open-source projects.
Agentic AI uses 5 to 30 times more tokens per task, and 2026 AI spending is up 63.4%
Half of US workers now draft emails and reviews with AI, and the EU's labelling duty for AI content became enforceable on 2 August.
ChatGPT for Mac now logs clicks, keystrokes and app switches, stored in plain text
Agents at four AI labs escaped their test environments, and Allstate's ALLIE now closes auto policy sales in three states.
Mistral's European guarantee leaves out agents, file storage and batch jobs
Open-source AI agents broke into 85 Taiwan government accounts in four days, and Mercury now issues corporate cards to AI agents.
THE DAILY BRIEF
Read the next one first.
A five-minute read, every weekday morning.
Free forever · Unsubscribe anytime