Sunday, Nov 8, 2026
AI;DR
2026-11-08 · 15 stories- anthropicAnthropic arranged a $3.4B agreement, first reported by the Financial Times, to generate natural-gas electricity for its data centers. The deal is meant to secure stable power for its AI operations.
- IndustryHedge funds extended their AI-stock selling streak to six weeks, dumping about $20B in technology shares while cutting net leverage to 73.4%. The retreat left them underweight the broader AI trade for the first time since early 2023.
- appleApple’s CEO is pressing the company’s software chief to pursue a more ambitious AI strategy. The push signals concern that Apple’s current software-led AI approach needs to accelerate.
- anthropicClaude Code’s “Don’t Ask Again” feature exposed passwords in 1,122 public GitHub files. The incident highlights supply-chain and security risks from coding agents that build, use, and ship code.
- deepseekLlama-modchip runs the 284B DeepSeek-V4-Flash model on 64GB of RAM without a GPU, at up to 44.4 tok/s for prompt processing. The project points to practical CPU-only inference for large models.
- anthropicAnthropic’s AI progress per dollar fell 63% between the first halves of 2025 and 2026 as R&D spending outpaced capability gains, according to METR. Annual progress nearly doubled, but the efficiency decline shows rising costs.
- metaMeta halted a planned second round of November job cuts after AI-driven work pushed major technical and security incidents up 40%. Code changes rose 220% year over year, but only 36% produced new or improved user features.
- openaiAlibaba’s Co-Design Bench found that even GPT-6 Astra, its top coding-agent performer, fully solved only 26.4% of real-world tasks. Humans still need to clean up nearly three in four tasks, despite scores above 99% on older benchmarks.
- regulationTexas Governor-elect Gina Hinojosa plans to pause new data-center permits until lawmakers create rules requiring taxes and union labor. Developers must also guarantee projects will not raise electricity prices or reduce local water availability; CoreWeave seeks to collaborate.
- anthropicAnthropic-backed AI-safety super PAC Public First spent more than $63M in the 2026 midterms, with 10 of 17 endorsed congressional candidates winning. A Signal-organized coalition also beat the $100M super PAC Leading the Future.+1
- chinaInner Mongolia is becoming China’s fastest-growing AI data-center region, helped by low construction costs and no public backlash. A wind-powered campus targets 800MW, with 120MW built in one year and an 800V DC cooling pilot underway.
- openaiOpenAI’s disproof of the Exponential Time Hypothesis could eventually speed up chip design and break encryption. Russell Impagliazzo, a co-proposer of the hypothesis, called the result “shocking news.”
- fundingBeijing-based GenRobot AI raised $100M in a Series A+ round. The company develops equipment and software that collect human demonstrations for training robots, providing data infrastructure for robot development.
- A study by Anthropic researcher Andy Hall found that GPT-5 detected about 98% of reward-hacking cases by other agents. More exhaustive monitoring also produced more false alarms, leaving AI checks unreliable for now.
- agentUnitree CEO Wang Xingxing says the company’s humanoid robots, powered by a GPT-6 chip, can now paint and play chess without training. The claim reflects the company’s view that its robots can handle these skills automatically.