Daily AI RoundupsSaturday, 15 August 2026

Nvidia's proposed financing guarantee for OpenAI's Ohio data center leads the 15 August edition after the chipmaker reportedly scaled the deal back to about $100 billion. Reuters reports that Washington plans to make partner countries choose between American and Chinese AI systems, while Anthropic disclosed more than $11.5 billion in preliminary second-quarter revenue and Nvidia reported large quarter-end stakes in SpaceX and Intel. The roundup also covers Qwen3.8-27B, Claude text watermarking, Anthropic's second Risk Report, US open-weight AI policy, Cloudflare's MCP security controls, unannounced OpenAI model constants and Matthew Green's analysis of AI-driven vulnerability discovery.

• AI infrastructure financing
• US-China AI policy
• AI economics
• open-weight models
• AI safety
• AI security
• agent infrastructure

Stories in this roundup

1. 01
   Nvidia nears a revised deal to guarantee about $100 billion of financing for OpenAI's planned Ohio data centerThe Information
2. 02
   The US plans to tell partner countries to choose American or Chinese AI systems, Reuters reportsReuters
3. 03
   Anthropic's preliminary second-quarter revenue exceeded $11.5 billion, up at least 14-fold year over year, with positive adjusted operating income, The Information reportsThe Information
4. 04
   Nvidia discloses quarter-end stakes worth about $21 billion in SpaceX and $30 billion in IntelBloomberg
5. 05
   Qwen releases the open-weight Qwen3.8-27B vision-language model with a native 262,144-token context windowQwen / Hugging Face
6. 06
   Anthropic details an upcoming Claude text watermark using SynthID-Text to comply with EU AI Act rules, without hidden characters or changes visible to readersAnthropic
7. 07
   Anthropic's second Risk Report raises its high-stakes misalignment assessment from 'very low' to 'low' and discloses a stronger internal Model 2Anthropic
8. 08
   Senator Jim Banks urges US incentives for domestic open-weight AI models and tighter restrictions on Chinese access to American chipsReuters
9. 09
   Cloudflare Gateway adds protocol-level MCP detection, a traffic dashboard, and controls to block connections that bypass approved portalsCloudflare
10. 10
    OpenAI's Go SDK adds model constants for unannounced Daybreak and gpt-5.6-cyber modelsOpenAI / GitHub
11. 11
    Johns Hopkins cryptographer Matthew Green argues AI bug hunting could shrink exploit stockpiles and renew government pressure for surveillance backdoorsA Few Thoughts on Cryptographic Engineering