Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 11,566|Added todayToday: 94

Category

Search

Today'sSafety & Ethics

ByteDance Creates AI Safety Team, Hires Globally

ByteDance’s Seed forms an Alignment & Preparedness team and begins recruiting 10 researchers globally
Positions are based in Beijing, Shanghai and San Jose, with resources comparable to frontier-model teams
Planned work includes safety training, vulnerability testing, agent test environments and independent research partnerships

Economic TimesRead more · 1 min →

Today'sSafety & Ethics

Anthropic Unveils Claude Distillation Defenses

Anthropic published defenses against unauthorized attempts to distill Claude’s capabilities.
Initial observations showed attackers collected about 90–99% less data.
The system detects and blocks reasoning-extraction requests, and can restrict or suspend repeat offenders.

Ledge AIRead more · 1 min →

Today'sHealthcare

Biohub Expands Virtual Cell Initiative to $1.8 Billion

Biohub expands its virtual-cell AI initiative with U.S. agencies, Google DeepMind, and other research organizations participating.
The collaboration totals $1.8 billion; Google DeepMind, Isomorphic Labs, and Meta are contributing $300 million combined.
DOE plans to invest more than $500 million over five years, while NIH will provide data supported by more than $500 million in prior federal investment.

Ledge AIRead more · 2 min →

primeNumber Launches Autonomous Data Platform

primeNumber began offering primeNumber ADP on October 15, 2026.
The service detects specification changes and failures, then proposes fixes for affected data, metrics, and workflows.
AI agents control the DWH, TROCCO, and COMETA through a management portal.

IT LeadersRead more · 1 min →

Vera Rubin NVL72 Posts First AgentPerf Results

Artificial Analysis posted initial Vera Rubin NVL72 AA-AgentPerf results against GB300 NVL72 across four models.
Vera Rubin reached up to ~5.7x throughput per megawatt on DeepSeek V4 Pro at ~190 tok/s per user.
The 72-GPU, 36-CPU system posted leaderboard highs of 906 tok/s on DeepSeek V4.1 Flash, 408 on GLM-5.2 and 255 on Kimi K3.

Artificial AnalysisRead more · 2 min →

Michigan Township Approves Google Data Center

Van Buren Township approved Google’s Project Cannoli data center site plan in a 5-1 vote on Oct. 14.
The 282-acre campus is projected to use up to 3.6 million gallons of water daily and 1 gigawatt of electricity.
County commissioners overrode a veto of $125 million in tax incentives, cutting certain property taxes by 50% for 12 years.

GovTech AIRead more · 1 min →

ITmedia Tests Claude Haiku 5.5 on Japanese Business Tasks

ITmedia compared Claude Haiku 5.5 with Claude Sonnet 5.5 and GPT-6 Luna on three Japanese business-writing tasks.
Haiku’s medium setting cost about one-seventeenth as much as Sonnet’s, with nearly equal scores on routine documents.
Haiku’s max setting averaged 160.7 seconds and ¥2.88, taking 12.6 times longer and costing 15.4 times more than medium.

ITmedia AI+Read more · 3 min →

Trending Keywords

Today'sRobotics

China Scales Up Humanoid Robot Production

Chinese manufacturers shipped 18,500 humanoid robots in the first half of 2026, surpassing 2025’s full-year global shipments.
Five factors support China’s production: government policy, concentrated funding, supply chains, orders, and on-site quality improvements.
Isao Sugena says Japan is strong in neural components and sensors but lacks a company mass-producing AI computing boards for robots.

ITmedia AI+Read more · 2 min →

SK hynix Shows Memory-Centric AI Rack

At OCP Global Summit 2026, SK hynix unveiled next-generation memory products and a memory-centric AI rack.
Its CXL 3.2-based 256GB CMM-DDR5 offers 80GB/s bandwidth, more than twice that of the previous generation.
MARS demos showed KV cache sharing and reuse to support long-context inference and improve memory efficiency.

SK HynixRead more · 3 min →

Today'sHealthcare

Agentic AI Connects Data Across Cancer Care

Snowflake describes agentic AI workflows across five oncology stages, from research and screening through survivorship.
A U.S. screening study found AI-supported mammography review identified 21.6% more breast cancers than standard review.
AstraZeneca cut data product development from six months with 16 engineers to four days with two using Snowflake Cortex AI.

SnowflakeRead more · 2 min →

SGLang Adds Nemotron 3.5 Super VL Support

SGLang added day-0 serving support for NVIDIA Nemotron 3.5 Super VL on October 15, 2026
The model has 120 billion parameters, activates 12 billion per token, and handles text, images, documents, and video
NVIDIA reports 243.9 tokens per second on RTX 6000 PRO and up to 5x throughput versus similar open models

LMSYSRead more · 1 min →

Researchers Introduce SWE-at-Scale Bug Synthesis

Shanghai Jiao Tong University researchers present SWE-at-Scale for generating software engineering tasks from historical bug fixes
The framework reverses real fixes into bug-induction examples and transfers learned bug patterns to novel code contexts
Authors report greater task diversity and realism than baselines, with bug diversity scaling through 200 generations per file

HuggingFaceRead more · 1 min →

NVIDIA, AI4Bharat Build Indic Corpus

NVIDIA and AI4Bharat introduced a 2.8 trillion-token corpus for 22 Indian languages with an open data-generation pipeline.
The corpus has 2,601.0 billion native-script tokens, 224.6 billion romanized tokens and 34 language-script configurations.
Nemotron 3.5 Super VL checkpoint scores rose 4.98 to 9.82 percentage points after an Indic continued-pretraining stage.

HuggingFace BlogRead more · 2 min →

Intern Lumina U2 Unifies Visual AI Tasks

Shanghai AI Laboratory researchers present Intern Lumina U2, a unified diffusion model for visual understanding and generation.
The 16.9B-parameter model activates 1.4B parameters and uses 8 discrete visual codes at each position.
The authors report a 4.45× image-generation speedup and leading scores across image, video and 3D benchmarks.

HuggingFaceRead more · 1 min →

JevAgentHarness Delegates Agent Tool Choices

University of Illinois researchers introduce JevAgentHarness to delegate agent tool choices to smaller decision models
AgentDecisionBench found the top model on isolated questions solved only 69.5% of agent tasks
AgentDecision-4B achieved 91.5% on unfamiliar tasks; trained models cut costs 81.1–89.0% and ran up to 2.5 times faster

HuggingFaceRead more · 1 min →

Sullivan & Cromwell Builds AI Agreement Analyzer

Sullivan & Cromwell and OpenAI developed Agreement Analyzer to review draft agreements using legal precedents and firm expertise.
The tool runs multiple review cycles, compares clauses and returns a memorandum with proposed tracked changes for lawyers to assess.
John Savva cited GPT-6 Astra’s reasoning advances as a way to support more legal research and writing.

OpenAIRead more · 1 min →

Yesterday'sInfra

Hornetsecurity Adds Microsoft 365 Security Capabilities

Hornetsecurity by Proofpoint announced Microsoft 365 security updates for AI readiness, with rollout planned in Q4 2026.
Tenant Risk Exposure flags six permission and sharing risks, while the interface provides remediation shortcuts and a tenant status report.
AI.MY brings adaptive Teams training with three tailored questions; a Control Panel agent offers product and security assistance.

PR NewswireRead more · 1 min →

YouTube Music Labels AI-Generated Tracks

YouTube Music begins labeling tracks created primarily with generative AI, with an explanation available by tapping the label.
Deezer plans a filter to remove AI tracks; it says more than 50% of uploads are AI-generated, with 90,000+ uploaded daily.
Spotify uses artist-profile badges, Tidal demonetizes AI tracks, and Apple Music plans voluntary transparency tags before year-end.

pcmag.comRead more · 1 min →

Yesterday'sSafety & Ethics

TASI 2026 Panel Debates AI Safety

TASI 2026 panel in New Delhi debated AI safety, child protection and regulation on October 14.
OpenAI, Google and Roblox representatives discussed red-teaming, risk sharing, age safeguards and India-specific safety evaluations.
Panelists called for safer defaults for children, stronger testing and clearer regulation as deepfake and privacy risks grow.

Indian ExpressRead more · 2 min →

Yesterday'sSafety & Ethics

Anthropic Defends AI Usage Oversight in Singapore

Anthropic co-founder Benjamin Mann defended usage oversight for powerful models at the opening of the company’s Singapore office.
Singapore ranks second among 121 places in Claude use per capita; more than 1,500 GovTech engineers use Claude Code, Anthropic says.
Mann warned publicly released open-weight models cannot retain resilient safeguards; Anthropic is working to host frontier models in Singapore.

FortuneRead more · 1 min →

Trending Keywords

Last 7 Days