Pao

Nous Research fine-tuned Alibaba's Qwen3.8-27B into Arete-27B, cutting the model's propaganda rate from 22 percent to 0.9 percent. Receipts: veriwire.news/daily-ai-news/…

Anthropic's own employees self-reported 12x productivity using its best internal model, and METR found its models now improve twice as fast as a year ago. None of 135 submissions from nine AI agents passed a new Carnegie Mellon–national labs benchmark for autonomous research; 114 failed integrity checks. Google will invest $17 billion in a Wyoming data center campus it bought unfinished from Crusoe; the sale price wasn't disclosed. Nous Research fine-tuned Alibaba's Qwen into Arete-27B, cutting its propaganda rate from 22 percent to 0.9 percent while general capabilities held steady. Baidu faces a securities fraud class action alleging it misled investors about AI offsetting declining ad revenue; the lead plaintiff deadline is November 13. 6 stories today, every one sourced → veriwire.news

Crusoe confirmed Friday that it sold its unfinished data center project in southeastern Wyoming to Google, which plans to invest $17 billion to develop the campus. Receipts: veriwire.news/daily-ai-news/…

METR's audit of Anthropic's internal research data found the pace of model improvement has roughly doubled since October 2025, with about 80 percent of the acceleration attributed to… Receipts: veriwire.news/daily-ai-news/…

Anthropic may have crossed its own AI acceleration warning threshold as early as July 15, METR found, with improvement roughly doubling since October 2025. Airbnb plans an AI travel agent for the first half of 2027, promising an interface much richer than a chatbot. Thinking Machines' Inkling 2 Small claims the top spot among western open-weight models, scoring 31 on the Artificial Analysis Intelligence Index. OpenAI will stop using non-disclosure agreements with public officials on data center projects, allowing only narrow protections for security and trade secrets. Physicl raised $10 million in seed funding to make simulations realistic enough to train robots, releasing Studio with 10,000 3D environments. 6 stories today, every one sourced → veriwire.news

OpenAI announced it will no longer use non-disclosure agreements with public officials on data center projects, allowing only narrow protections for security information, trade… Receipts: veriwire.news/daily-ai-news/…

Airbnb said in its Q3 2026 shareholder letter that it plans to launch an AI travel agent in the first half of 2027 to help users plan and book trips. Receipts: veriwire.news/daily-ai-news/…

Phonon ranks first in German and Spanish intelligibility on Hugging Face's Open TTS Leaderboard, placing in the top three across all evaluated languages against models up to 100 times… Receipts: veriwire.news/daily-ai-news/…

Claude Code's product lead, Catherine Wu, has left Anthropic for OpenAI, where she'll head product management for Core Products. Discovered Materials, a YC-backed startup, is putting $1 million behind five unsolved materials science problems frontier models haven't cracked. A free on-device text-to-speech model, Phonon, ranks first in German and Spanish intelligibility against models up to 100 times its size. Together AI says its new agent-driven compiler re-optimizes inference as traffic changes, with self-reported gains up to 2.6x. Google Research released Agile Deliberation, a human-in-the-loop framework for subjective image classification, averaging 10.5 F1 points above zero-shot Gemini 2.5 Flash. 6 stories today, every one sourced → veriwire.news

Muhammad Tufail Khan has released Bansat, a blueprint-first framework and Agent Skill for Claude, available via GitHub. Receipts: veriwire.news/daily-ai-news/…بنسټ-lay-the-foun-248614/

Starting in early 2027, Google will require schools to accept AI for students or lose access to Google Search, as The New York Times first reported. Receipts: veriwire.news/daily-ai-news/…

Starting in early 2027, schools that turn off Google's AI will lose Search too — Google says the two are too fused to separate. Nobel physics laureate Geoffrey Hinton expects AI to trigger a large-scale incident, like a financial or power blackout, within two years. Gemini trains on your chats even if you pay — and chats read by human reviewers are kept up to three years after deletion. Tencent's WeChat Vision team released WeVisOne-VL, an open vision-language model in 2B and 4B sizes unifying detection, OCR, segmentation and matting. 4 stories today, every one sourced → veriwire.news

Anthropic reports that message threads reduce Claude's median time-to-first-token by 35% for large requests. Receipts: veriwire.news/daily-ai-news/…

Bug hunters can now earn up to $172,000 from OpenAI after it quadrupled its bounty payouts. Huawei is telling AI chip customers to source their own memory, with smuggled HBM hitting roughly $4,000 per unit. Anthropic says Message Threads cut time to first token by 35 percent on the longest requests; two outlets covered the same launch. Anthropic added rules stopping Claude Code from cloning existing games after a wave of Opus 5.5-built knockoffs. A transformers bug let UMT5's decoder attend to future tokens during training, sitting in main until September 28. 7 stories today, every one sourced → veriwire.news

The Letter Company has launched a Letterpress-1 beta, a coding agent that converts site and market signals — including Google rankings — into sandboxed changes submitted as pull… Receipts: veriwire.news/daily-ai-news/…

OpenAI plans to merge the Chat and Work tabs within ChatGPT, according to an email sent to users. The change will initially reach select users for feedback before a wider rollout. Receipts: veriwire.news/daily-ai-news/…

$2.6 billion is what Saudi Aramco is paying for a 49 percent stake in AI firm Humain, plus its own AI technology. OpenAI is merging ChatGPT's Chat and Work tabs, starting with select users before a wider rollout. HelpPuff is a free, open-source website assistant that runs entirely inside your own Cloudflare account after one prompt. Taghash launched Brain, an AI assistant that turns fund-management tasks which took days into minutes. 4 stories today, every one sourced → veriwire.news

A Vals AI study testing hundreds of political questions found Chinese AI models show a lower-than-expected "pro-China lean," though answers in Simplified Chinese skewed more… Receipts: veriwire.news/daily-ai-news/…

Chinese AI models show less pro-government bias than expected on political questions — though answers in Simplified Chinese still skew that way, a Vals AI study of hundreds of questions found. Google, Stanford, and Oxford researchers built StreamVibe-VL, a video agent that knows when to speak up during tasks, beating other models by 10.7 percent. 2 stories today, every one sourced → veriwire.news

According to the Wall Street Journal, OpenAI has raised roughly $11 billion toward a $30 billion round, falling short of both its funding and valuation goals. Receipts: veriwire.news/daily-ai-news/…