From a $9.1B compute lock-in to an IPO runway and an ethics exit — Anthropic dominates today's AI cycle, while Meta makes a credible return to open-weight models.
Anthropic signed a 20-year, $9.1B compute agreement with Riot Platforms, securing 191 MW of dedicated power at its Rockdale, Texas campus. Riot's stock jumped ~25% on the news.
Sources tell the WSJ that Anthropic is in early investor discussions ahead of what could be a record-breaking public listing, touting rapid revenue growth and plans to address mounting public backlash against AI.
Meta released Muse Glimmer, a 30B-parameter open-weight multimodal model optimised for local agentic workflows, benchmarking competitively with models twice its size and shipping with day-zero Hugging Face support.
OpenAI released GPT-5.6-Cyber, a security-tuned model that can autonomously scan codebases and network configurations to surface exploitable vulnerabilities, gated to enterprise security teams and not consumer access.
Anthropic announced that new Claude models deployed in the EU will add watermarks to text outputs and C2PA metadata to generated files, complying with the EU AI Act transparency obligations that took effect August 2. Older models will be updated.
Chloé Bakalar has departed as OpenAI's head of ethics after under a year in the role, following recent exits by the head of safety systems and a former head of mission alignment.
Google DeepMind published results showing WeatherNext outperforms ensemble numerical weather models on tropical cyclone track and intensity simultaneously — a combination that has resisted single-model solution in operational meteorology.