AISep 2, 2026Ars Technica
Google releases Gemini 3.8 Flash, its third Flash model in six weeks
Ars reads Gemini 3.8 as cadence covering for a missing Pro: three Flashes in six weeks, OSWorld still trailing Claude Opus, Cyber as a niche.
The file
Latest
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
AISep 2, 2026Ars Technica
Trump may be forced to reveal secret rules feds use for AI safety testing
Protect Democracy is suing four agencies for the unclassified AI safety-review framework Trump's White House will describe but will not publish.
AISep 2, 2026NVIDIA Developer
The Modern CUDA Toolbox in Practice: A Step-by-Step Optimization Walkthrough
Nvidia's developer blog turns a broken 6.8-second image pipeline into 23 milliseconds, and the moral is stop writing the kernel CUB already shipped.
AISep 2, 2026TechCrunch
US government sides with OpenAI on issue of training LLMs on copyrighted material
TechCrunch's shorter take is the procedural one: a 20-page brief invoking last year's AI executive order, filed in a court the White House does not run.
AISep 2, 2026The Verge
Researchers fear safety disaster ahead of OpenAI’s Astra release
The Verge's Astra story is the quote: Ryan Greenblatt calling a more opaque architecture maybe the worst safety development to date — and OpenAI not confirming the architecture.
AISep 2, 2026Google DeepMind
Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
DeepMind's Gemini 3.8 post is a vendor scoreboard: same introductory price as 3.7, more tokens if you let it 'work harder,' and Cyber locked behind Fairwind.
AISep 2, 2026The Verge
The Trump administration is supporting OpenAI in the NYT copyright lawsuit
Washington just walked into the Times-OpenAI copyright fight on the lab's side, arguing that a licensing default would choke 'American prosperity.' That is policy dressed as fair-use doctrine.
AISep 2, 2026NVIDIA Developer
Co-Designing AI Models Using Speculative Decoding for Faster LLM Inference
NVIDIA's third co-design note is not a product launch. It is a serving manual: pick draft length and draft mechanism by where you sit on the throughput-latency curve, or you will pay for tokens nobody accepts.
AISep 2, 2026WIRED
The Logical End Point of AI Job Interviews Is Two Bots Talking to Each Other
A contractor put ChatGPT on speaker with an AI recruiter after five dead-end screens. The bots complimented each other. No human ever called back. This is the hiring stack eating itself.
AISep 2, 2026The Verge
Google is sending MrBeast into the wilderness, armed with AI
Google hired YouTube's first 500-million-subscriber channel to take Gemini camping. This is not a research drop. It is a bid to make the model look useful to people who do not read model cards.
AISep 2, 2026TechCrunch
HiddenLayer nabs $100M as enterprises rush to secure their AI deployments
HiddenLayer raised $100 million as Gartner says AI-security spend will hit $2.83 billion this year. The market is real. Whether runtime AI security stays independent is not.
AISep 2, 2026The Verge
OpenAI accused of ‘aiding and abetting’ Tumbler Ridge mass shooting in dozens of new lawsuits
Thirty new California filings accuse OpenAI of aiding the Tumbler Ridge shooting, not just failing to stop it. That is a harder claim — and a fight over who inside the company got to say no to the police.