Capital keeps flowing to AI labs and infrastructure despite the post-midterm selloff: Thinking Machines Lab is reportedly about to close a $5 billion round at a $40 billion valuation, and Groq is pitching investors at $8 billion. On security, Apiiro found 2,539 secrets leaked on GitHub, most of them in Claude Code's approval file. Nous Research's Arete-27B shows that state-aligned behavior in Chinese open models can be stripped by fine-tuning alone, while Microsoft has disbanded the AI Frontiers group behind AutoGen.
According to Business Insider (Nov 6), Accel is leading a $5 billion round for Mira Murati's lab at a $40 billion pre-money valuation (about $45 billion post-money), which could close within days. That is more than triple its $12 billion 2025 seed valuation but below the $50 billion discussed in late 2025, showing investors are still backing frontier labs despite market turmoil.
Bloomberg (Nov 6) reports that executive chairman Alex Davis is pitching a Series B at $8 billion, more than double the $3.15 billion August valuation. After its $20 billion licensing deal with Nvidia, Groq shifted from chips to running inference data centers, and it pitches itself as less debt-laden than CoreWeave or Nebius as neocloud stocks sell off.
Apiiro found that when users pick 'Yes, and don't ask again', Claude Code saves the full approved command, including passwords or tokens, to .claude/settings.local.json; 1,122 public copies contain secrets, for 2,539 secrets across 1,994 repositories in total, including 379 database passwords and 23 AWS keys. Of 16 coding agents reviewed, Claude Code is the only one that writes full commands to a project file by default, a lesson in the risks of granting agents standing permissions.
Arete-27B is a supervised fine-tune (about 53 million tokens) of Qwen3.8-27B, released under Apache 2.0 with its training data and an 803-prompt eval; the propaganda rate fell from 22% to 0.9% with general capability nearly unchanged. The result suggests state alignment in Chinese open models is a thin layer, which matters for companies that want strong but more neutral open models.
In a Nov 6 post, John Langford said Microsoft Research's roughly 50-person AI Frontiers group, creator of the AutoGen agent framework, was moved into Microsoft AI in September and is now 'over'; he and the other founding leaders have left. It is another example of Big Tech AI research groups being folded into units focused on large-scale model building.
TechCrunch (Nov 7) reports that US voters rejected data centers in dozens of local ballot measures, many restrictions or moratoriums passed, and Texas's governor-elect campaigned on limiting data centers. The pressure directly threatens compute expansion plans at LLM labs such as OpenAI and xAI.
If these digests help you make sense of technology, a small contribution keeps NHỊP picking out what is worth reading.
Scan the QR code with a Vietnamese banking app.
Download QR ↓ · View full image ↗