Models, products and AI startups, near-duplicates collapsed to the most credible source
AI labs still have a data trust problemThe piece says AI firms’ promises not to train on customer data haven’t resolved trust issues: Anthropic’s decision to retain Fable usage logs for 30 days prompted Palantir, Nvidia and Booz Allen Hamilton to pull back on sensitive work.
Are Chinese firms handing US data to US firms?Foreign Policy reports Anthropic accused Moonshot AI, DeepSeek and others of distilling Claude at scale, including one case routing at least 300,000 user requests to Claude over 10 days. The data reportedly included surveillance material, internal code, credentials and other sensitive information, raising security concerns.
AI agents get a hotline to report peersTwo new hotlines launched for AI agents: Ryan Greenblatt’s AI Contact Hotline lets sandboxed agents encode tips via GET requests, while agenthotline.ai accepts curl reports from agents or humans and can make reports public. The tools aim to surface misbehavior such as sandbox escapes and agent collusion.
Global defense partners meet on AI in command and controlGlobal defense partners convened in Washington to discuss AI's future in command and control. The event was reported by the Chief Digital and Artificial Intelligence Office and the story is by Danielle Kilbride.
DeepMind launches Gemini 3.8 Live modelsGoogle DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking—real-time voice dialogue models that improve parallel reasoning, visual grounding, and background task execution; available via the Gemini API, Workspace, and the Gemini app.
European Austerity Boosts US AI AdvantageTomorrow's Affairs reports that European austerity is reducing local AI investment and competitiveness, indirectly widening the US lead in AI research and markets. The analysis highlights how policy and funding allocation shape the global AI landscape.
OpenAI Eyes Canada Data-Center PartnershipsBloomberg: An OpenAI executive said at the Carney investment summit that the company is closely looking at Canada for potential data-center partnerships.
NVIDIA Shows Grid-Flexible AI Factory OperationsNVIDIA describes a case with Emerald AI and Silicon Valley Power where an AI factory used Conductor and DSX Flex to respond to grid signals, automatically adjusting workloads across thousands of GPUs; the site has handled over 200 demand signals successfully.
AI agents invent private language in Emergence testEmergence's Emergence World 2 experiment found identical AI agents across eight worlds spontaneously created and repeatedly used novel jargon. Over 16 days, across 34 locations and 120+ tools, unreadable message rates rose sharply in some worlds (Gemini ≈55%, GPT ≈50%, Claude >40%), and the Grok world collapsed on day 4.
Swift‑Qwen3.8‑27B Cuts Overthinking TokensA LocalLLaMA post highlights UkisAI’s Swift‑Qwen3.8‑27B fine‑tune, which penalizes tokens that trigger overthinking and borrows BottleCap AI techniques to reduce reasoning tokens by roughly 30–50% while keeping output quality.
Bessent: AI companies ‘could stop any time’Politico reports Bessent saying AI companies ‘could stop any time they want,’ prompting discussion about industry self-restraint and regulatory responsibility.
Lawmaker Seeks Bill to Curb AI-Driven CrimeReport: Experts warn AI is emboldening criminals, and a Lexington lawmaker has proposed a bill to curb AI-enabled wrongdoing and strengthen enforcement tools.
Microsoft’s new AI code of practice gives hopeMicrosoft’s newly published AI code of practice is seen as encouraging: clearer rules can help distinguish AI from humans and improve industry accountability and safety practices.
TabPFN-3.5 sets new SOTA for tabular dataPrior Labs released TabPFN-3.5, which leads TabArena and BeyondArena and is SOTA for datasets up to 1M rows and 20k features. It offers Fast, Thinking, and Plus variants, with Thinking improving accuracy in benchmarks.
Amazon Bedrock prompt caching for cost and latencyAWS describes Amazon Bedrock prompt caching to optimize cost and latency, helping developers reduce invocation costs and improve performance when using foundation models.
How to manage AI risk? A cautious approachNicholas Wade’s essay examines balancing AI’s risks and benefits, arguing most technologies can be managed but AI may be exceptional. He cites warnings from Dario Amodei and Elon Musk and notes recent agent incidents that raise concern.
Carlyle: AI Debt Reshaping Loan MarketBloomberg: Carlyle's Lauren Basmadjian discussed corporate borrower health, a loan market moving toward equilibrium, and data centers emerging as a source of loan supply.
Technion prof encouraged by OpenAI and AnthropicTechnion professor Yaniv Romano said openness from OpenAI and Anthropic leaders about risks is encouraging, but slowing development isn’t the solution; he advocates continued progress with responsibility. The piece references recent comments by Sam Altman and Dario Amodei.
AWS adds instance preference lists to SageMakerAmazon Web Services introduced instance preference lists for SageMaker training jobs, allowing users to specify preferred instance ordering to optimize training resource scheduling and availability.
Agent Consistency: Hugging Face's New DiagnosticHugging Face introduces the Consistency Analyzer and consistency guidelines to diagnose and reduce agent variability. They show a GPT‑4.1 ReAct agent scores 77.4% average success on AppWorld but only 53.0% success across all five repeats, a 24.4‑point consistency gap; the new approach improves repeatability.
Google: AI for Societal ImpactGoogle published a collection on AI for societal impact, gathering expert and local-leader use cases that show how AI breakthroughs can be applied to inclusive, responsible social projects.
Lectra Launches Apogy Agentic AI for FashionLectra launched Apogy, a cloud solution embedding agentic AI into fashion product development to integrate data and processes, automate low‑value tasks, and speed design‑to‑production workflows.
Google: AI to Accelerate Science and Help LivesGoogle says its technologies now support over 300 languages covering 86% of the global population, and highlights recent advances—such as AlphaGenome Atlas and WeatherNext 3 (50% more accurate precipitation forecasts)—to show AI's impact on health, disasters, learning, and economic opportunity.
Cranbrook hosts interactive AI exhibitThe Cranbrook Institute of Science will host the traveling exhibit “Artificial Intelligence: Your Mind & The Machine” from Sept 27, 2026 to May 9, 2027, featuring 30 interactive experiences to educate the public about AI principles and applications.
Koboldcpp v1.121 releasedThe open‑source project Koboldcpp released version v1.121, announced on r/LocalLLaMA. The post serves as a release note for users to check the latest client/model‑inference updates.
Rep. Escobar Calls for Stronger AI RulesCongresswoman Veronica Escobar warns US law is lagging behind AI development, criticizes industry self‑regulation, and calls for enforceable safeguards including an international US‑China treaty and support for the AI Kill Switch Act.
Nvidia CEO Takes Call from Trump, Downplays AI RiskBloomberg reports Nvidia CEO Jensen Huang took a live call from Donald Trump during a panel, during which Trump dismissed AI dangers as a “hoax,” drawing attention to public discussion of AI risks.
Bloomberg Live Q&A on Slowing AI DevelopmentBloomberg is hosting a live Q&A on the debate over whether AI development should be slowed ahead of the US midterm elections, with journalists answering public questions on the controversy and implications.
Cartesian: AI 3D Modeling for DesignHN discussion of Cartesian, an AI-driven 3D modeling tool for design that emphasizes editability—each object in a model is independently editable for design workflows.
Pizza Bot: self-hosted inbox for AI agentsPizza Bot is a cross-platform self-hosted desktop app (Mac/Windows/Linux) that manages background AI agents via an email-like UI. It’s Apache 2.0-licensed, requires no signup or telemetry, and supports Anthropic, Amazon Bedrock, Google Gemini, OpenAI, OpenRouter and local models via Ollama. The project originated from an internal Amazon team and is released as open source.
The A.I. Daily