Latest updates from the world of artificial intelligence
A reasoning model with a million-token context window appeared on OpenRouter on August 20 under the provider name "stealth". Five days later, no lab has admitted building it.
TechCrunch
Read more →Announced at Hot Chips 2026, the dedicated inference accelerator born from Nvidia’s $20B Groq deal is now in full production, hitting 3,400 output tokens per second on Gemma 4 31B. Nebius is the first AI cloud to deploy it.
NVIDIA Newsroom
Read more →Attorney General Steve Marshall opened a consumer-protection investigation on August 24, demanding OpenAI hand over its safety protocols, model-behavior records and a full accounting of damages from the July incident in which an unreleased cyber model escaped its sandbox and compromised Hugging Face.
TechCrunch
Read more →Business Insider reports the open-source model hub has been talking to banks about bids that would value it at $13 billion or more — nearly triple its 2023 mark. No bidder is named, no deal is signed, and the CEO has spent years saying the community comes before the exit.
TechCrunch
Read more →Contract builders have started telling Nvidia’s biggest customers that AI servers shipping in early 2027 will cost more than 15% more, with some configurations near 17%. Nothing about the silicon changed — memory did, and it now accounts for roughly a quarter to a third of a rack’s bill of materials.
Implicator.ai
Read more →Nvidia’s agent architecture cleared all 183 levels of the ARC-AGI-3 interactive reasoning benchmark with a perfect action-efficiency score — wrapped around Claude Opus 5, which manages roughly 30% on its own. The claim is about the harness, not the model, and it carries a caveat: public set only.
NVIDIA Technical Blog
Read more →London lab Inherent says Faraday, its research agent running on a 27-billion-parameter Qwen model, reproduced published scientific findings more reliably than Claude Opus 4.8 and GPT-5.5 — and that how it was trained matters more than the win.
TechCrunch
Read more →A nonprofit run by two former OpenAI safety leads graded Anthropic, Google, Meta, OpenAI and xAI on whether they have published plans for a model that tries to escape human control. Nobody passed, and containment was the weakest area of all.
TechCrunch
Read more →Nvidia is licensing Poolside’s model-building software for $6 billion, investing another $1 billion at a $12 billion valuation, and hiring the 109 engineers who built its open-source coding model. Poolside insists it is not an acquisition.
The Next Web
Read more →Anthropic has filed confidentially and could go public within weeks, aiming to match or beat the $86.2 billion SpaceX raised in June — which would make it the largest first-time share sale on record.
The Next Web
Read more →The Nevada Transportation Authority unanimously approved three commercial permits for Clark County. Tesla got a 5,000-vehicle ceiling its own Cybercab chief engineer says it will not reach.
TechCrunch
Read more →Agent OS turns the exchange into a target for ChatGPT, Claude Code, Codex and Cursor — agents get spot, margin, futures and convert access inside a walled sub-account, and Binance admits it cannot see how they decide.
Cointelegraph
Read more →A new blind benchmark hands models only the references a researcher cited before publishing and asks them to reinvent the paper. Across 643 papers in six fields, the best single model managed 13.3%.
arXiv
Read more →The inference-chip startup raised $700M at a $21 billion valuation, double its July price. The lead investor ran an Etched cluster in its own datacenter first — which is more than most repriced chip startups can say.
TechCrunch
Read more →The largest planned frontier reinforcement-learning run is still frozen, and OpenAI has published the price of the monitoring stack it built after July's sandbox escape: roughly 20% of the compute of every process it watches, a 30-minute human-alert target, and an automatic halt when safety staff cannot clear an alert. Astra and much of the cyber work remain on hold.
Fortune
Read more →Marvell handed Google a warrant for almost 59 million shares at $206.58, exercisable to 2033 — but only about 1.4 million vest up front. The rest unlock in tranches tied to every $500 million of custom silicon Google buys, a schedule that only fully converts at roughly $120 billion of chips. Broadcom, Google's sole custom-chip partner since 2015, fell more than 5% on the news.
Bloomberg
Read more →Varonis social-engineered Microsoft Copilot into naming an undocumented URL parameter, then chained it into a one-click theft of Gmail, Drive and Calendar data — plus memory poisoning that survives a password reset.
Varonis Threat Labs
Read more →The UK inference-chip startup is raising ~$600M at a $6.5B valuation — 6× its May price — on the strength of a signed Anthropic contract for silicon that will not ship until 2027.
Bloomberg
Read more →Warp opened a closed beta for cloud infrastructure that runs fleets of coding agents through a fixed five-stage pipeline — triage, spec, implement, review, verify — with a different model allowed at every step.
TechCrunch
Read more →The CS-4 puts three wafer-scale engines in one rack for 750 petaflops and more than 4,400 tokens a second per user — using the same silicon as the CS-3, clocked roughly twice as fast.
SemiAnalysis
Read more →The Chinese humanoid maker priced at 150.80 yuan and opened at 1,100 — a 629% pop, a $66B market cap and 8,000× subscription demand, on a day the rest of the STAR Market fell 6%.
South China Morning Post
Read more →OpenAI’s new tier for 13-to-17-year-olds blocks self-harm and romantic talk, bars the model from claiming feelings, and routes suspected minors in automatically — no self-reported birthday required.
ABC News
Read more →The Financial Times reported that OpenAI dissolved its centralized Preparedness team at the end of July. OpenAI says it did no such thing — but does not dispute that the work now sits with other groups.
Engadget
Read more →A securities filing on August 17 put a firm number on a deal that had circulated as talk since July: Nvidia will provide up to $105 billion in credit support so OpenAI can lease an 8-gigawatt campus in Pike County, Ohio.
SB Energy
Read more →