On mvanemmerik.com, Nov 16: NFL Week 10 chalk talk: Detroit dazzles, Cincinnati takes the North
vanemmerik.aiClaude, ChatGPT and Grok news, twice a day

Sunday, November 8, 2026

Anthropic reportedly pays $3.4 billion for 1.7 GW of gas generators to power its data centers

Anthropic is reported to have taken over $3.4 billion in Caterpillar generator orders from OnePWR Solutions, in what the FT calls the largest known power-equipment buy by an AI lab. Anthropic also published a study of about 207,000 research sessions showing scientists lean on Claude mostly for data analysis and writing. OpenAI's partnership with security firm Hacktron and its BioBasicsBench wet-lab benchmark round out a quiet Sunday; nothing new from xAI.

Sunday, November 8, 2026

ClaudeBusiness

Anthropic reportedly pays $3.4 billion for 1.7 GW of gas-fired generators to power its data centersReported

The Financial Times reports that Anthropic agreed to pay about $3.4 billion for 1.7 gigawatts of heavy-duty gas generators, plus engineering and installation work, by taking over orders that San Antonio-based OnePWR Solutions had placed with Caterpillar about two years ago. The commitments were reportedly secured in late September; Anthropic did not buy OnePWR, and it has not commented on the report.

Why it matters: If accurate, it would be the largest known single purchase of power equipment by an AI lab, showing how far labs are going to secure electricity outside the grid queue.

AnthropicHappened Nov 8, 2026 · Sources: Yahoo Finance (Investing.com, citing the FT) · FourWeekMBA

ClaudeResearch

Anthropic study of 207,049 research sessions finds scientists mostly use Claude for data analysis and writing

A report from Anthropic's Economics team sorted 207,049 consumer research sessions from Aug. 2 to Sept. 30 (128,914 in Claude, 78,135 in Claude Code) by field and task, and surveyed 528 scientists. Data analysis made up 33.7% of Claude sessions and 48.1% of Claude Code sessions, against 27.6% of researchers' time, while writing was 32.0% of Claude sessions. Anthropic models that wide adoption of today's capabilities could speed research up by 1.35x to 2.02x depending on the field, with lab-bound sciences gaining least.

Why it matters: It is one of the larger looks at how working researchers actually use AI tools, though the speed-up figures are modeled scenarios rather than measured results.

ClaudeClaude CodeAnthropicHappened Nov 5, 2026 · Sources: AI TL;DR · FourWeekMBA

OpenAISafety

OpenAI partners with Hacktron, the security firm whose July report chained a forum bug into OpenAI's internal code

OpenAI announced on Nov. 6 that its engineers and researchers at Hacktron, a startup building AI penetration-testing agents, will jointly pick high-risk areas to test, starting with how OpenAI's systems connect to third-party services. Hacktron also gets wider access to OpenAI's cyber models through the Daybreak Red program. The two first worked together through OpenAI's bug bounty, which paid Hacktron $6,500 for a July report on a chain from a community forum to account takeover and an internal repository.

Why it matters: OpenAI is moving from paying for one-off bug reports to standing, model-assisted red-team work on its own systems, weeks after Reuters reported internal security warnings.

OpenAICodexHappened Nov 6, 2026 · Sources: Unite.AI · ExplainX

OpenAIResearch

OpenAI and Ligo publish BioBasicsBench, 100 routine wet-lab tasks where GPT-6 Astra scores 66.7%

OpenAI and enzyme-design company Ligo introduced BioBasicsBench on Nov. 6: 100 tasks built from raw lab evidence such as gel images, plate photos, microscopy and sequencing reads, scored with partial credit. In OpenAI's own tests GPT-6 Astra led at 66.7%, ahead of GPT-6.1 Sol at 62.0% and GPT-5.6 Sol at 43.8%, while using about half GPT-5.6 Sol's output tokens. OpenAI says a higher score does not yet mean a model can make reliable lab judgments, and no other lab has reported results.

Why it matters: It measures everyday lab interpretation rather than textbook biology, an area where model errors such as miscounted colonies still show up.

OpenAIHappened Nov 6, 2026 · Sources: Unite.AI · Winzheng

ClaudeResearch

Anthropic details how Claude 'dreams' between sessions to tidy agent memory, lifting a wiki test from 13% to 48%

In a Nov. 6 research post, Anthropic describes dreaming, where Claude reviews past transcripts between sessions to fix wrong notes and write new ones. In its own tests, a wiki Claude built over 128 dreaming sessions from about 350,000 internal Stainless files raised scores on 25 questions from 13% to 48%, and in Pokémon Red, Claude Fable 5 kept about 10 memory files and earned four badges, where Sonnet 3.7 wrote 413 files and earned one. Anthropic notes the wiki helped less when a plain text search could find the answer in one document.

Why it matters: Memory that corrects itself is one of the main limits on long-running agents, and these are early, mostly single-run results from Anthropic itself.

Try it: Dreaming is a research preview in Claude Managed Agents on the Claude Platform; developers request access from Anthropic.

ClaudeClaude APIAnthropicHappened Nov 6, 2026 · Sources: Let's Data Science · Interesting Engineering · Claude blog: New in Claude Managed Agents