AI news4
Pentagon adds ChatGPT and Grok to its GenAI.mil platform, with Anthropic still absent
The Department of Defense announced on 1 September that GenAI.mil, its secure AI platform for handling sensitive but unclassified information, now includes custom versions of OpenAI's ChatGPT and xAI's Grok for Government alongside Google's Gemini, which launched on the platform last year. More than 1.7 million of its 3 million-plus personnel have already signed on. Anthropic's Claude remains missing from the bundle, a direct consequence of the company's ongoing dispute with the Pentagon over contractual guardrails against mass surveillance use.
techcrunch.com ↗Why it matters
three of Anthropic's biggest rivals are now embedded across a 3-million-seat government deployment while Anthropic's own access dispute plays out in court.
Europe commits 387.8 million euros to an AMD-powered AI supercomputer
The EuroHPC Joint Undertaking signed a 387.8 million euro contract with French vendor Bull on 31 August to build LUMI-AI, a supercomputer powered by AMD's Instinct MI430X GPUs and EPYC processors rather than Nvidia hardware. The system will be installed at a data center in Kajaani, Finland, and should deliver ten times the AI capacity of today's LUMI system when it comes online in 2027.
hpcwire.com ↗Why it matters
a nearly 400 million euro sovereign AI-compute commitment built on non-Nvidia silicon is a concrete step in Europe's push to reduce dependence on a single chip supplier.
Sales-data platform Clay raises a new round at a $7 billion valuation
Clay, an AI-powered sales and marketing data platform, is closing a new funding round led by Wellington Management at a $7 billion pre-money valuation, Axios reported on 31 August. That is up from the $5 billion valuation the company set as recently as January 2026.
axios.com ↗Why it matters
a fast-rising valuation for an AI-native sales-operations platform, not a model lab, shows investor appetite for AI extending well beyond foundation-model companies.
Alibaba open-sources a preview of its next-generation Qwen4 architecture
Alibaba's Qwen team released Qwen3.8-Flash-Next on 26 August, a multimodal model with 125 billion total parameters but only 6 billion active per token, built to preview architecture choices planned for the full Qwen4 model family. The release includes a novel 51-billion-parameter embedding layer the team says is one of several innovations slated for Qwen4.
marktechpost.com ↗Why it matters
giving developers an early, open preview of a next-generation architecture months before the full model ships is a distinct strategy from simply releasing a bigger model, and it signals confidence in the underlying design.
AI in the nonprofit sector1
Sentient Futures kicks off its Fall 2026 Project Incubator for AI and animal welfare
Sentient Futures, an organisation working at the intersection of AI safety and animal advocacy, began its 10-week Fall 2026 Project Incubator on 31 August, pairing fellows with mentors on projects spanning AI and animal welfare, artificial sentience, and AI governance. More than 80 mentors from partner institutions are involved, and the program is open to anyone committed to improving outcomes for sentient beings, with fellows dedicating at least 5 hours a week through 9 November.
sentientfutures.ai ↗Why it matters
this is a concrete, funded structure turning AI-and-animal-welfare from a conversation into a pipeline of mentored projects, directly in the vein Kyle's own animal-advocacy work sits in.