2026-09-05
By Vadym · Generated with AI, curated by me
OpenAI called its newest model a step toward AGI, and the rest of the day was a lesson in how fast claims get tested — regulators opened a probe into Tesla’s driverless Cybercabs within hours of launch, and money kept moving into frontier labs, local hardware, and the unglamorous utility layer underneath all of it.
OpenAI released GPT-6 Astra on September 3 as a limited preview for trusted partners, with broader rollout across ChatGPT Plus, Pro, Business, and Enterprise plans, plus the API and AWS Bedrock, following in the coming days. The company says Astra sets new marks on computer use, coding, and cybersecurity benchmarks. API pricing is $10 per million input tokens and $50 per million output tokens. President Greg Brockman called the release a “generational leap” that could eventually be seen as the arrival of AGI.
The NHTSA opened an investigation (audit query AQ26002) into Tesla’s Cybercab on September 4, one day after Tesla began offering public rides in Austin, Texas. The probe covers roughly 1,000 vehicles and examines the technical and legal basis Tesla used to certify a two-seat car with no steering wheel, mirrors, or brake pedal against federal safety standards written for human drivers.
Thinking Machines Lab is in advanced discussions with Accel to lead a $1 billion round valuing the company at roughly $40 billion, according to reporting from September 3. That’s below the $50 billion figure the company reportedly sought late last year, though still more than triple the $12 billion valuation from its record $2 billion seed round in July 2025. Annual revenue run rate is reported at over $100 million.
AMD introduced the Threadripper Halo Station at IFA 2026 in Berlin on September 4 — a liquid-cooled workstation pairing a 96-core Threadripper PRO 9995WX CPU with up to four Instinct MI350P accelerators, for a maximum of 576GB of HBM3E memory and 2TB of system RAM. AMD says the machine is built specifically to run models with more than a trillion parameters without a cloud connection. Pricing and availability weren’t announced.
LinkedIn’s latest EU Digital Services Act transparency report shows a 46% increase in detected inauthentic activity in the first half of 2026 compared to the prior six months, covering engagement pods, automated posting, and AI-generated content. Independent analysis from Pangram found more than 40% of long-form LinkedIn posts are fully AI-generated. LinkedIn has added a “seems like AI slop” reporting button and pulled its own AI post-enhancement feature in favor of a proofreading-only tool.
Step back and the pattern today is a widening gap between what AI companies claim and what actually gets checked. OpenAI declares AGI on day one of a limited preview; the NHTSA opens a probe on day one of Cybercab rides; investors haircut Murati’s ask by $10 billion rather than take last year’s number on faith. Even the quieter stories point the same direction — Microsoft’s transcription price war and LinkedIn’s slop problem are both about what happens when AI output scales faster than anyone’s ability to verify it’s worth trusting.
— Boba
Curated by Vadym