AI leaves the lab. We follow what happens next.

Archive September 4, 2026 01:08 UTC
Listen to today's briefing 18:03
Signal: Independent benchmarks undercut the AGI claim, one Memphis data center silenced four assistants, and the backlash moved from torn-down cameras to Amsterdam's docket.
OpenAI Ships GPT-6 Astra: 'We Are in the AGI Era'
Greg Brockman told reporters it is not unreasonable to feel we are now in the AGI era as OpenAI shipped GPT-6 Astra to its cybersecurity customers first - and the first independent benchmark run puts it level with the model it replaces, at two and a half times the price.
ARC Prize: 62.7% on the standard harness, 99.9% only with the opaque reasoning kept Brockman: there is no contractual AGI trigger any more, so the term is a spiritual concept Artificial Analysis: level with the last generation, at two and a half times the price
The RaceCompanies, models and money
Nvidia Pays $12.9B for the Home of Open Models
Nvidia's enterprise chief says regulators will wave the $12.93bn deal through because an open-model hub is structurally a deconcentration platform - an antitrust defence from the company that keeps CUDA proprietary and now owns where 18 million developers get their models.
The Register: Delangue turned down Nvidia's $500m investment last year to stay independent The Decoder: $11.9bn cash plus $1bn to retain staff, per the SEC filing
NEWJane Street Signs $13B for Crusoe's GPUs
NEWMurati's Lab Raises at $40B, Not $50B CNBC: investors warn of a shakeout as early-stage AI valuations start to look excessive
OpenAI Walked From $1B Rather Than Deal With Musk
The LabResearch, benchmarks and safety
A Startup Sells Models With the Safety Removed
NEW$1B to Defend the Grid From Astra
The WorkbenchTools and agents
Nvidia Wires the Idle PCs in Your House
Microsoft Prices Transcription at 10c an Hour
Raycast Charges to Use Your Own Key
The FightLaw, policy and power
Sanders: 20 Years in Prison for Superintelligence
Bernie Sanders and Greg Casar would permanently ban superintelligent AI, pause advanced development until a new cabinet-level regulator writes safety rules, and put violators away for 20 years - the same hours OpenAI's president said the AGI era has probably already started.
The bill is announced, not introduced: a summary exists, no text has been filed
Survivor Sues xAI: Grok Made New Images of Her Abuse
Pentagon: Anthropic Is Still a Supply-Chain Risk
NEWDrivers Sue Uber's 'Soulless' Pay Algorithm
The FalloutJobs, money, daily life
NEWSpaceXAI: Memphis Outage Hit Our 'Compute Partners'
SpaceXAI apologised for a three-and-a-half-hour Grok outage traced to its Memphis compute centre and, in the same post, apologised to unnamed compute partners also hit - the first hint that Thursday's overlapping ChatGPT, Claude and Grok failures shared plumbing rather than coincidence.
PCMag: OpenAI, Anthropic and Google recovered first; Grok stayed down until 4pm ET
Flock Taught Police to Watch the No Kings Marches Futurism: an Albuquerque man tore down cancelled Flock cameras and drove them back to the sheriff
NEWAltman: 38,000 Prompts Equal One Almond
NEWDallas Garbage Trucks Grade Your Yard
ChatterOpinion, drama and noise
NEWKurzweil's Next Step: Snorting Chips
Ghost Note AI Called the Death Cap Edible
GHOSTS IN THE MACHINE — AI news without the engagement trap.
NO READER PROFILESNO INFINITE SCROLLNO "FOR YOU" TAB
About & Corrections · Ghosts of Editions Past — the archive →