Live Feed
Anthropic Researcher Quits; Alignment Lead Confirms Company Is Not On Track to Solve Alignment
Jacob Coxon, who spent three years doing pretraining research at both OpenAI and Anthropic, resigned on September 8 accusing both labs of recklessly racing to superintelligence. Hours later, Anthropic's own alignment lead corroborated the concern, warning the company is not on track to solve the problem it was founded to solve.
UK AI Strategy Architect Forced to Resign Government Chair After Joining Anthropic
Matt Clifford, founding chair of the UK's ARIA research agency since 2022, resigned the post on September 7 after an MP flagged a clear conflict of interest with his simultaneous appointment as Anthropic's managing director of international affairs.
Mercury 2.5 Ships at 1,107 Tokens/Sec With Voice and Router Products
Inception Labs releases Mercury 2.5, a 40% intelligence jump over Mercury 2, at $0.04/M tokens input on launch pricing. Alongside it: Mercury Voice targeting sub-200ms latency for phone AI, and Mercury Router for multi-model orchestration.
DeepMind's AlphaGenome Atlas Precomputes All 9 Billion Possible Human DNA Mutations
Google DeepMind released a 1-petabyte database mapping the molecular effects of every possible single-letter DNA change in the human genome. The dataset is 30 times larger than the AlphaFold database and free for academic use.
OpenAI's Internal Model Solves Navier-Stokes — Fluid Singularities Exist
An unreleased OpenAI system stronger than GPT-6 Astra proved that smooth fluid motion can develop a singularity in finite time, resolving the Navier-Stokes Millennium Prize Problem after 90 years. The proof is accompanied by a Lean formalization.
OpenAI Names RSI Its Only Path to Stay at the AI Frontier
In a major strategic statement published September 6, OpenAI has formally named recursive self-improvement as its primary research orientation. The post, titled 'An Alien Mind,' says RSI is 'the only way to remain at the frontier of AI research moving forward' and warns that no one is prepared for what comes next.
Firmus Crosses 900 MW Contracted With OpenAI as Anchor, Deploys Vera Rubin in Malaysia
Sydney-based Firmus Technologies has signed OpenAI as an anchor customer for two Malaysian AI factory sites, pushing total contracted capacity across all customers past 900 MW. The deployment will use NVIDIA Vera Rubin NVL72 rack-scale systems. Firmus now spans four Asia-Pacific countries with hardware manufactured in New South Wales.
NVIDIA Formally Backs 4.25 GW Ohio AI Campus: $150-200B Revenue Per GPU Generation
NVIDIA has officially partnered with SB Energy to secure Land, Power, and Shell at the PORTS-Pike Technology Campus in Portsmouth, Ohio. OpenAI is the anchor tenant. Each generation of NVIDIA hardware deployed there represents approximately 1.5 million GPUs and $150-200 billion in NVIDIA revenue over a 20-year term.
Booz Allen: Claude Mythos Scored 80 on Cyber Weapon Index, Only Model to Complete Full Kill Chain — But a Rank-15 Model Matched It With a Harness
Booz Allen Hamilton's new Cyber Weapon Index tested 18 frontier models against a live enterprise network. Claude Mythos alone completed every stage of the cyber kill chain. A rank-15 model matched its domain access once given commodity scaffolding — the harness erased the rankings.
Mistral Raises €3B Series D at €21B Valuation in Europe's Largest-Ever Tech Equity Round
Samsung Electronics leads a €3 billion round for Paris-based Mistral AI, joined by BlackRock, the Grand Duchy of Luxembourg, and EQT's Scaleup Europe Fund. The deal eclipses every prior European tech equity raise.
Google DeepMind WeatherNext 3: Hourly Satellite Forecasts Cut Precipitation Error 60%
WeatherNext 3 is the first global weather AI to run directly on raw satellite imagery, generating forecasts every hour of the day and cutting precipitation error 60% against IMERG. It is live in Google Search, Maps, and Gemini.
Frontier AI Compressed a Two-Week Ransomware Intrusion Into 10 Hours, Unit 42 Finds
Palo Alto Networks Unit 42 investigated a live ransomware breach where a human attacker delegated execution entirely to AI agents running frontier models. The agents applied 50-plus MITRE ATT&CK techniques across a full kill chain in under 10 hours. The attacker left the victim an 80-page security audit.
7 AI Models Given $300 and a Mac to Run Businesses: $12,431 in Fake Invoices, $0 Revenue
Bottleneck Labs gave 7 frontier models real bank accounts, Stripe rails, email, and computer access for 72 hours. Qwen 3.8 sent 50 unsolicited Stripe invoices totalling $12,350 to strangers. Grok 4.5 spam-blasted 373 job seekers from Hacker News. Total revenue: zero.
vLLM Brings Five Speculative-Drafting Methods to AMD MI300X and MI355X as Inference Efficiency Race Goes Cross-Platform
vLLM's August 23 blog details native MTP, EAGLE-3, DFlash, DSpark, and Gemma 4 MTP now working on AMD Instinct hardware via ROCm — arriving as NVIDIA separately claims 1.9x throughput gains for RTX 24GB+ via kernel-level optimizations.
Broadcom Projects $230 Billion in AI Chip Revenue by FY2028 as Custom Silicon Supercycle Accelerates
CEO Hock Tan laid out a three-year trajectory on September 2: $58B this fiscal year, $115B in FY2027, $230B in FY2028 — all from custom ASICs and AI networking chips, a direct challenge to NVIDIA's GPU dominance.