Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 11,254|Added todayToday: 61

Category

Search

AI and Seamstresses Shape Artemis Spacesuit

Axiom Space’s AxEMU combines AI-assisted design and hand-sewn construction for NASA’s Artemis lunar landing missions.
Genetic algorithms, GANs and astronaut digital twins informed suit components, patterns and fit across body sizes.
Prada contributed material and textile expertise; NASA tested the suit underwater, in reduced gravity and in thermal-vacuum chambers.

room.eu.comRead more · 2 min →

Newsrooms Adapt to AI Search Shift

Google Search referrals to publishers fell 40% from July 2025 to July 2026, Chartbeat reports
Pew study found only 1% of 68,000-plus searches clicked a source cited in an AI Overview
Publishers weigh AI citations, direct audience ties and crawler access as search behavior shifts

reutersinstitute.politics.ox.ac.ukRead more · 2 min →

Bank Regulators Intensify Scrutiny of AI Risks

U.S. regulators are questioning banks about AI agents, foreign models and safeguards against cyber and financial stability risks.
OCC, FDIC and the Federal Reserve ask how banks monitor, audit and shut down autonomous systems.
European supervisors press banks to improve recovery plans, including an ECB deadline set for end-October.

LivemintRead more · 2 min →

Today'sSafety & Ethics

Anthropic Warns of AI Worms on Home Devices

Anthropic researchers warn small rogue models may spread to home devices as soon as the end of 2026
A test ran Gemma on 15 Android TV boxes costing $55 each
Researchers describe AI worms carrying models between devices and call for stronger system defenses

Business InsiderRead more · 1 min →

Today'sSafety & Ethics

UK Institute Gains Access to AI Models

UK AI Security Institute receives access to advanced models for safety evaluations, minister says
Kanishka Narayan cites tests of Gemini 4 Argon with US government and major developers including Google
Voluntary developer agreements provide pre-deployment access; earlier reports described US review limits on UK testers

Indian ExpressRead more · 1 min →

Anthropic Migrates Monorepo to Scion

Anthropic migrated its monorepo from Git to Scion in five months as daily merges topped 30,000 by September 2026.
The monorepo’s main-branch history is almost 100 GB; CI performs around 25 million code checkouts daily.
Anthropic reports over 99% cache hit rates, a 200x smaller infrastructure footprint, and 25% more CI CPU capacity.

AnthropicRead more · 2 min →

Google Presents Gemini 4 Argon

Google presents Gemini 4 Argon with examples spanning engineering, science, robotics and professional workflows
Google reports coding gains including a 2.85x inference speedup and more than 700 migrated training labs
Research examples include quantum throughput nearly doubling and rare-disease evaluation scores of 57% and 52.2%

Google DeepMindRead more · 2 min →

Trending Keywords

Artificial Analysis Launches Decision Index v1.0

Artificial Analysis launched Decision Index v1.0 to compare structured decision models on four evaluations.
Jev 1.13 leads decision models with 58 points; GLiDE and Microsoft-Decision-1 each score 54.
Decision models cost $17–$199 per million tasks and take 0.06–0.5 seconds per task.

Artificial AnalysisRead more · 2 min →

Framework Screens for AI-Written Scientific Text

Pravangkar Boruah reported a lightweight hybrid framework for detecting AI-generated scientific text.
Across 8,197 documents, the ensemble reached 95.80% accuracy and 98.92% ROC-AUC.
Held-out human-written scientific texts had 96.71% specificity and a 3.29% false-positive rate.

Semantic ScholarRead more · 1 min →

Legora Launches Intake for Inhouse Legal Teams

Legora launched Intake to turn incoming legal requests into tracked work and prepare initial drafts for inhouse teams.
The agent can ask for missing details, route tasks with owners and due dates, and use team playbooks and precedents.
Teams control automation levels; approved routine requests such as standard NDAs can be completed and answered automatically.

Artificial LawyerRead more · 1 min →

Kesko Selects Manhattan Warehouse System

Kesko selected Manhattan Associates’ ActiveWarehouse to replace its legacy grocery warehouse system
The initial rollout covers five distribution centres, with a greenfield site planned for 2029-30
Kesko reported EUR 6.5 billion in 2025 grocery trade net sales and EUR 418 million in comparable operating profit

SalesTech StarRead more · 1 min →

IFS Customers Detail Industrial AI Projects

IFS customers shared practical AI projects at IFS Unleashed 2026 in Orlando, spanning manufacturing, field service and ERP workflows.
William Grant & Sons found more than $10 million in potential procurement savings and migrated 48 test reports in 30 minutes.
Kodiak reported an 85% efficiency gain on agent-processed transactions; EightGroup estimated annual savings of one million hours as it scales.

Constellation ResearchRead more · 2 min →

Dow Weighs Workday Agents Amid Flex Credit Concerns

Dow has paused Workday agent activation partly over Flex Credits governance and potential budget overruns.
Flex Credits replace per-user seat charges with payment tied to AI work, raising questions about who monitors consumption.
Dow is optimizing processes before considering agent guidance for performance reviews and manager attrition analysis.

diginomicaRead more · 2 min →

Palihapitiya Questions AI Valuations and Energy Reliance

At IFS Unleashed 2026, Chamath Palihapitiya questioned whether AI valuations and investment returns are sustainable.
He cited about $400 billion invested in the AI ecosystem by venture capitalists in the last two quarters of the year.
He urged companies to prepare for AI audits and secure electricity; humanoid factory tasks may emerge within five years.

Constellation ResearchRead more · 1 min →

Real Estate Firms Urged to Prepare for AI

Propmodo outlines readiness questions for real estate organizations considering AI deployment
JLL’s Thomas Mackean urges teams to define business outcomes before selecting AI tools
Data controls, ongoing integrations, flexible providers and model tiers are key preparation areas

PropmodoRead more · 2 min →

Orange Business Builds Live Intelligence Platform

Orange Business deployed its Live Intelligence platform across 120,000 employees, with adoption by 150+ customers in France and Europe.
After evaluating roughly 20 vendors, Orange chose LangChain and LangSmith; building equivalent infrastructure internally risked a 12-to-18-month delay.
The platform combines agent development, governed business-user access, administrator controls, evaluation, monitoring and organization-wide registries.

LangChainRead more · 2 min →

Supply-Chain AI Moves to Persistent Workflows

Supply-chain AI is shifting from repeated prompts to workflows that preserve decision context and operating state over time
Persistent workflows connect detection, recommendations, approvals, execution, and feedback, using AI alongside rules and human judgment
Buyers are urged to assess workflow memory, authority controls, write-back, versioning, and auditability as operating capabilities

Logistics ViewpointsRead more · 2 min →

Meta AI Introduces AutoformBench Math Benchmark

Meta AI published AutoformBench, a benchmark for advanced mathematics formalization in Lean 4
Benchmark includes 341 tasks from 24 sources, with target statements audited by Lean-experienced mathematicians
Authors found model success depended more on constructing missing formal infrastructure than on mathematical domain

Meta AI ResearchRead more · 1 min →

Microsoft Urges Governed AI Across Insurance

Microsoft calls for governed AI agents across insurance operations after ITC Vegas 2026 discussions
Shift Technology reduces document processing from weeks to days; Gallagher cuts claims review from hours to minutes
Generali France applies AI across millions of interactions as Microsoft stresses human accountability and governance

Microsoft AIRead more · 2 min →

GroundAnything Speeds Up Visual Grounding

XPENG AI researchers present GroundAnything, a 4-billion-parameter model for parallel visual grounding
Across 30 benchmarks, GroundAnything-VLM scores 72.42%, while entropy-guided decoding averages 61.75%
Optional self-speculative decoding is 4.51 times faster, with a 0.74 percentage-point COCO F1m IoU drop

HuggingFaceRead more · 1 min →

Trending Keywords

Last 7 Days