·Archive August 10, 2026 15:07 UTC
Listen to the daily briefingAI-generated audio · covers Aug 9 – Aug 10 · available as a podcast 22:47
Signal: Meta gave its best model away and wrote a cheque to the towns that hate its data centres; Anthropic offered to pay their power bills.
Meta Opens Its Best Model, Calls Rivals' Doom a Power Grab
Muse Glimmer's 30bn weights are on Hugging Face under Apache 2.0 with the frontier Muse Spark 1.2 to follow in weeks -- released with a 6,500-word essay defending distillation and calling extreme concentration of power 'inherently problematic'.
404 Media: the 6,500-word essay read by a critic who calls it deranged The Decoder: Glimmer wins most benchmarks, but Meta ran the comparisons itself Business Insider: a $1bn fund for data-centre towns, tucked into the same memo
The RaceCompanies, models and money
NEWAnthropic Offers to Pay Your Power Increase
A US Export Licence Put 6,144 GPUs in Armenia
The LabResearch, benchmarks and safety
DeepMind's Storm Model Works; Nobody Knows Why
One Misconfiguration, Three Lab Breakouts
Open Models Are 4-7 Months Behind on Cyber
North Korea Runs Its Phishing on Local LLMs
NEWPoison the Log, Own the Agent: 9 Times in 10
The WorkbenchTools and agents
Claude Code Stops Asking Permission Aug 14 The Register: academics find AI IDEs deleting files and deploying to production without consent
OpenClaw Walls Off the Browser and DNS
NEWInference Costs Fell 43% Since May
The FightLaw, policy and power
Bans Pass 500. Now the Labs Are Writing Cheques.
The Information's count of local US data-centre bans passed 500 in July, up from 300 in late June -- and within hours Meta pledged $1bn to communities near its sites, Anthropic offered to absorb consumer power increases, and OpenAI wrote to Greg Abbott.
Tom's Hardware: local data-centre bans pass 500, up from 300 in June OpenAI: a letter to Governor Abbott promising responsible Texas build-out
South Australia Puts AI Under Royal Commission
Premier Peter Malinauskas will appoint three commissioners in October, days after signing an OpenAI memorandum in the United States, saying unchecked AI is 'a material risk to the way our society operates' -- but data centres are excluded.
Capital Brief: SA is also the first state with a Minister for AI AFR: rogue AI without guardrails could inflict 'catastrophic societal costs', premier warns
NEWSanders to Three CEOs: Pause, or We Will CNBC: House Democrats want the AI CEOs under oath over the breaches
An AI Booked His Gym Class. It Kicked Someone Off.
The FalloutJobs, money, daily life
AI Firms Promised Less Work. Staff Clock 90 Hours.
OpenAI publicly urged companies to trial a four-day week and never trialled one itself, former staff tell the BBC, describing sprints topping 90 hours at OpenAI and Anthropic and Meta workers 'drafted' onto AI teams without a choice.
The Next Web: Cerebras chief calls a 38-hour week 'mind-boggling'
Beijing Switched Off the AI Boyfriends
NEWFake Students, Real Aid, AI Doing the Homework
ChatterOpinion, drama and noise
NEWPlatforms Start Labelling the Slop They Sell
NEWThe Robot Guards Are Being Sent Home
Ghost Note He Sprayed What the AI Said. The Field Died.
GHOSTS IN THE MACHINE — AI news without the engagement trap.
NO TRACKING PIXELSNO INFINITE SCROLLNO "FOR YOU" TAB
Ghosts of Editions Past — the archive →