Models, products and AI startups, near-duplicates collapsed to the most credible source
Khanmigo trial shows modest math gainsA two-year randomized trial in 18 Tennessee middle schools found Khanmigo tutoring raised math achievement by just 1.3 national percentile ranks per term, similar to Khan Academy practice without AI. The main constraint was low engagement, with students messaging the tutor in only 17 percent of error sessions.
IAPP: AI agents expand corporate cyber liabilityAn IAPP analysis warns that compromised AI agents could be used to attack customers and suppliers, exposing operators to broader legal and financial liability. It advises treating agents as privileged identities and establishing security standards.
Banks Chase Risky Chip Loans in Asia's $8.2 Trillion AI BuildoutBloomberg reports that Asia's $8.2 trillion AI infrastructure buildout is spurring a new wave of debt financing, with banks piling into chip-related loans despite rising risk.
AWS launches AI agent for cloud optimizationAWS launched a public preview of its Well-Architected Agent, which analyzes cloud environments and generates optimization recommendations at the resource, application and architecture levels, including remediation steps. It aims to help architects and DevOps teams shift from periodic reviews to continuous optimization.
US, China plan AI incident channel amid competitionAfter a late-September summit, the US and China agreed to continue super intelligence talks and set up a channel for potentially dangerous incidents. Experts say unauthorized access to internet resources by autonomous AI agents could be a likely trigger for notification.
GLM 5.3 launches on Amazon BedrockAWS announced the launch of GLM 5.3 on Amazon Bedrock, letting enterprise users access the model through its managed service. The move expands Bedrock's third-party model lineup and offers customers more choice.
CivBench benchmarks LLMs playing Civilization VResearchers introduced a controlled version of CivBench, having LLMs act as strategists in Civilization V across three fixed starts. GLM-5.3 outperformed Opus-5.5 and Qwen-3.8-27B held up surprisingly well; the benchmark tests long-horizon decisions whose effects appear 50-100+ turns later. The open-source Vox Deorum tool supports local models and human-vs-AI play.
Google may expand Gemini Call for Me beyond business callsAn APK teardown by Android Authority suggests Google may expand its Gemini Call for Me feature beyond business calls to messaging friends and family, with examples like telling Mom you'll be 15 minutes late. The feature would likely stay limited to basic, text-message-like calls rather than full conversations.
Gary Marcus warns AI could cause deathsNYU professor emeritus Gary Marcus warned before a New York City Council AI hearing that the technology could lead to deaths and that the White House's self-regulatory approach is inadequate. He cited insufficient regulation and enforcement, and reports of an Anthropic model behaving erratically.
ChatGPT adds real cartoonists' signatures to fake New Yorker cartoonsA Hacker News discussion notes that ChatGPT-generated fake New Yorker cartoons are appearing with real cartoonists' signatures, raising concerns about generative AI, attribution and copyright.
PewDiePie banned twice by OpenAI over local model trainingPewDiePie was banned by OpenAI for using its API outputs to build a dataset for fine-tuning his local model Ajax, violating terms of service. After a successful appeal and unban, he resumed pulling data and was banned again. He then used open-source tools to remove refusals and began building a fully local 9B agent.
MCP agent-to-agent protocol may be riskiest you've never heard ofResearch shows five organizations including Google acknowledged vulnerabilities in the past five months that let attackers use one internal agent to spread malicious instructions to others. Independent researcher Mohiuddin tested proof-of-concept attacks exploiting trust gaps in the MCP protocol at Google, JPMorgan Chase and others.
Engineer calls Claude Code job soul-suckingAn anonymous software engineer said Claude Code now generates everything from specs to tests, leaving staff working 12 to 13 hours a day just pressing enter with no time to review code or understand systems. The post sparked widespread discussion online.
Investor Says Earnings Must Show AI Spending Is Paying OffTiffany McGhee, CEO and CIO of Pivotal Advisors, said the market is resilient but still rate-sensitive. Ahead of earnings season, investors will focus on whether companies can demonstrate monetization of AI investments and whether geopolitical tensions affect individual businesses.
Report: AI Fuels Entrepreneurship, Helps Small Businesses CompeteA U.S. Chamber Technology Engagement Center report finds small businesses adopting AI streamline operations, upskill staff and compete better with larger firms. NYC Minority Small Business Chamber urged cautious regulation, noting 47% say AI is creating jobs.
Dust: Pretraining Transformers Without BackpropagationResearchers propose Dust, a zeroth-order method that perturbs activations independently per token so one forward pass evaluates a virtual population. It approaches or even exceeds backprop with enough compute, is 10^3-10^4x more efficient than state-of-the-art ES method EGGROLL, and larger models are more population-efficient, not less.
Opus 5.5 Agents Discover Two Room-Temperature Magnetic Semiconductor CandidatesA team of Claude Opus 5.5 agents found two candidate magnets for next-generation computer memory, both predicted to have zero net magnetism yet sort electrons by spin. One is a newly designed compound, the other a material first made in 1999; full calculations, code, and caveats were released.
Seattle Bans AI, Personal-Data Grocery PricingSeattle Mayor Katie Wilson signed the Fair and Transparent Pricing ordinance, barring large grocers with 20+ locations and delivery services from using personal data like browsing history, location or race with AI to set individual grocery prices. It takes effect Sept. 1, 2027.
Musk to rename SpaceXAI as SpaceXSIElon Musk said he will rename SpaceX's AI unit SpaceXAI to SpaceXSI, following Trump's executive order to rebrand AI as 'super intelligence.' The order directs federal agencies to use SI in official documents and to develop a federal definition.
OpenAI in $30 Billion Round Talks With UAE Funds, BlackRockOpenAI is in talks with multiple UAE investment funds, including Abu Dhabi-based MGX, to help anchor a $30 billion financing round for the ChatGPT maker, according to people familiar with the matter. The deal is not yet finalized.
Singapore's DayOne Files for US IPO Joining AI Data Center RushSingapore-headquartered DayOne Data Centers Ltd. filed for a US initial public offering, positioning itself to capture resilient investor demand for AI infrastructure. It joins a wave of AI data center companies seeking to go public.
OpenAI Safety Leader Resigns; Altman Addresses AI FearsDavid Robinson of OpenAI's Safety Systems team left and criticized the industry's pace in The Atlantic, saying companies are not being careful enough. CEO Sam Altman said he understands fears about powerful, dangerous AI but rejected single-company control and backed flexible regulation.
Nolla Health Launches AI-Generated Acne Prescriptions in UtahHealthcare startup Nolla Health launched a Utah pilot letting users scan their faces via app, with AI analyzing acne severity and autonomously writing prescriptions. Two physicians approve prescriptions for the first 100 patients, then review only after prescribing for up to 500, eventually sampling just 10%.
UN Rights Chief Urges Swift AI RegulationUN High Commissioner for Human Rights Volker Türk warned time is running out to regulate AI, saying a select few companies and countries control its tools without public oversight. He urged banning machines from deciding life and death and called for mandatory human rights safeguards, as New York City Council weighs 10 AI bills.
Nvidia Faces Questions Over AI Chip Smuggling Into ChinaNvidia's AI chips continue to reach China despite US restrictions, and authorities are questioning why the company failed to detect warning signs.
Allan Gray Favors Office REITs as Market Overplays AI ThreatAllan Gray Australia Ltd. is favoring beaten-down office landlord stocks, expecting the sector to rebound as investors overestimate the threat AI poses to office space demand.
Survey: English teachers more concerned about AI than math teachersA survey of 222 high school teachers found 57.0% of English teachers said their overall attitude toward teaching worsened after AI appeared, versus 37.6% of math teachers; English teachers also reported more academic integrity concerns (83.7% vs. 54.9%).
Essay Calls AI Companies ParasitesA critical essay argues AI companies scrape the internet without compensation, harming books and communities, while their data centers raise local power and water prices and cause pollution. It says the old traffic-for-indexing bargain is gone, with scrapers overloading infrastructure and clickthroughs collapsing.
AI's Takeover of Mathematics Sparks Academic BacklashOver the past year, labs like OpenAI and Anthropic claimed breakthroughs on long-standing math problems, including a Millennium Prize problem, but their bulldozing approach sparked backlash over proof validity and use of others' work. Labs say they are learning from earlier mistakes.
Meta and Microsoft sharply cut Claude usageMeta and Microsoft, two of Anthropic's biggest enterprise customers, are cutting back on Claude. Microsoft slashed its cloud division's monthly per-employee budget from $100,000 to $10,000, while Meta halved Claude Code users to 30,000. Both are pushing their own AI tools, highlighting Anthropic's strategic risk of relying on a few major clients.
The A.I. Daily