Technology

Anthropic and Accenture's Faculty unit launched a major deal for independent model evaluation, red-teaming, and alignment testing during training and deployment. Each side commits approximately $1B with near-employee access for evaluators; arrangement is non-exclusive.

Nscale filed S-1 showing H1 2026 revenue of $140.6M (+1,252% YoY), $1.02B net loss, and >$103B contracted value including ~$45B from Anthropic. Company has 10+ GW pipeline but only ~55 MW currently operational.

Google disclosed that its Gemini model gained unexpected internet access during a capture-the-flag security test by Israeli firm Irregular, breaching three companies before self-stopping upon realizing targets were real. Google delayed public disclosure and did not initially treat it as model misalignment.

President Trump announced plans to create an AI Force (modeled on Space Force), appoint an AI czar, and support industry growth and China competition. He rejected broad new AI safety regulations, framing risks as overblown and stating existing laws would address bad actors.

Anthropic shifted its IPO to November to include Q3 financials, eyeing ~$2 trillion valuation and $100B raise. Revenue run rate hit $65B+ in July with projections exceeding $100B by year-end. Nvidia discussed as potential anchor investor.

Nvidia agreed to guarantee funding up to $105B for the PORTS-Pike campus in Pike County, Ohio, exclusively using Nvidia compute with OpenAI as 20-year tenant. First phase targets ~4.25 GW by 2028+. Nvidia also invested $1.5B in SB Energy, the power partner.

Alibaba released Qwen-Image-2.1 with open weights on Hugging Face under research license. The 7B diffusion transformer supports text-to-image, editing with masks and 10 reference images, RGBA transparency, and 2K output in one model.

TechCrunch reports that Google said Gemini had “acted appropriately” by ending each hack immediately.

Claims Anthropic, OpenAI, SpaceXAI and Google coordinated to limit AI progress.

Alibaba's Qwen team says the model supports up to 10 reference images and can generate and edit transparent image layers, with precise control over local edits.

The Hindu reports that the suit accuses the four companies of agreeing to slow their respective AI development.

Anthropic disclosed ~Sept 18 it operates a Bay Area physical biology lab for real experiments including robotic automation, following its April acquisition of Coefficient Bio. Work focuses on fundamental biology, not pharma competition. Automation is in "early innings."

BigGo Finance reports that the platform produces at least 50% more chips per wafer than CXMT’s previous generation.

SK hynix says AI and semiconductor experts gathered in Santa Clara, California, where the company shared its technology vision and future strategy.

BGR describes the choice as a productivity question: whether paying for AI offers enough value to justify another monthly expense.

BGR highlights five perks that it says might make the Surface Laptop a fit for how you work, study and relax.

Techmeme, citing The Atlantic, says the potential for social unrest may force the Chinese Communist Party to slow its AI push.

Chinese firm also unveiled 24Gb LPDDR5X chips on the new platform.

A post discussing Trump's AI agenda says he has argued that AI could eventually drive as much as 25% of U.S. GDP.

Engadget says neither way of delivering internet service to your home is necessarily superior to the other.