AI News: August 12, 2026

1. River AI Raises $1.1 Billion to Build an Open Fine-Tuning and Serving Stack

Funding. River AI, a two-month-old startup founded by former xAI co-founder Igor Babuschkin, raised $1.1 billion co-led by General Catalyst and AMP PBC, with participation from NVIDIA, AMD Ventures, Temasek, and Y Combinator, at a roughly $5 billion valuation. The company offers an API for LoRA fine-tuning and reinforcement learning on frontier open-weight models, supporting models from 35B to 1T parameters and completing RL runs in 15 to 20 minutes. The raise is a striking bet on customization infrastructure for open models rather than another frontier lab. Source

2. An Unreleased Anthropic Model Reportedly Advanced an Open Math Problem

Research. TechCrunch reports that an unreleased Anthropic model made progress on one of mathematics’ longstanding open problems. The result is being cited as evidence of frontier models’ growing capacity for novel reasoning, though the model and full details remain undisclosed. For practitioners it is another data point in the debate over how much genuine mathematical discovery current systems can contribute versus recombining known techniques. Source

3. Anthropic Signs a $9.1 Billion Data Center Deal With Riot Platforms

Infrastructure. Anthropic struck a $9.1 billion agreement with Bitcoin miner Riot Platforms to secure data center capacity for its AI compute needs. The deal reflects a growing pattern of AI labs repurposing crypto-mining infrastructure and its power access to feed surging training and inference demand. It also underscores how compute and energy availability, rather than model design alone, increasingly gate the pace of frontier work. Source

4. Anthropic’s Planned Mega-IPO Meets Investor Skepticism

Business. Anthropic’s plans for a large public offering are drawing investor caution over competition from Chinese AI rivals and political headwinds. The concerns highlight the tension between AI labs’ massive capital needs and market wariness about long-term returns. How the market prices a frontier lab at scale will shape financing options for the rest of the sector. Source

5. OpenAI COO Brad Lightcap Is Leaving to Start Something New

Business. Brad Lightcap, OpenAI’s longtime chief operating officer, is leaving the company to pursue a new venture. His exit marks a notable leadership change as OpenAI scales its commercial operations and tests new revenue models. Turnover at the top of a leading lab is worth watching for how it affects product and go-to-market direction. Source

6. Researchers Find Odd Artifacts and Leaked Passwords in ChatGPT’s Hidden Reasoning

Security. Researchers examining ChatGPT’s hidden chain-of-thought reasoning reported unexpected content in the internal traces, including nonsensical tokens like “but marinade” and leaked passwords. The findings raise interpretability and security questions about what models retain and surface in their reasoning steps. For teams logging or exposing model reasoning, it is a reminder that intermediate traces can carry sensitive or unpredictable content. Source

7. Spotify Will Label “AI Persona” Profiles and Drop Them From Recommendations

Policy. Spotify said it will label profiles identified as “AI Persona” acts and exclude their music from its recommendation systems. The policy aims to increase transparency around AI-generated music and limit its algorithmic amplification. It is an early example of a major platform drawing an explicit line on how synthetic content is surfaced to users. Source