Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 12,417|Added todayToday: 13

Category

Search

SGLang Reports DeepInfer Benchmark Results

SGLang reports prefill throughput up to 1.04x DeepInfer on 4x GB300 and 4x B300
On 64x GB300, SGLang decode steps were 1.1% to 7.4% shorter at 16 to 256 requests per GPU
SGLang integrates DeepSeek kernels and adds KV caching, prefill/decode separation and replica load balancing

LMSYSRead more · 2 min →

Anthropic IPO Prospectus Details Vast Spending Risks

Anthropic’s draft IPO prospectus describes $80 billion in first-half losses and a possible $2 trillion valuation.
Anthropic plans $750 billion in computing capacity spending and $450 billion in cloud agreements over seven to 10 years.
CalPERS is cutting its Magnificent 7 holdings as Meta raises its stated infrastructure spending to about $825 billion.

independent.co.ukRead more · 2 min →

Today'sSafety & Ethics

Anthropic Verifies 123 Open-Source Libraries with Claude

Anthropic and Theorem formally verified security properties in 123 OSS-Fuzz libraries, proving behavior across 239 test entry points.
The work found hundreds of bugs, including CVE-2026-56132, and completed proofs totaling 180 million lines.
Opus 5.5 solved half of 36 verification tasks using about 40% fewer tokens than Opus 5.

AI Times KoreaRead more · 3 min →

Google Cloud Unveils Agent Compute Engine

Google Cloud announced a compute service that gives each AI agent its own microVM.
The system handled 200,000 agents and 500 activations per second, with suspend and resume taking under 500 milliseconds.
Agent Substrate supports up to 10 times the computing density and adds enterprise governance controls.

AI Times KoreaRead more · 2 min →

32 Banned Nvidia B300 GPUs Detected on China Telecom Network

Israeli security firm Lavee detected 32 export-banned Nvidia B300 GPUs on China Telecom’s network.
Lavee identified the GPU models through exposed DCGM Exporter data, but could not locate the chips physically.
Reports have previously described banned Nvidia chips entering China; Nvidia says the detected batch is tiny compared with millions of authorized GPUs.

AI Times KoreaRead more · 1 min →

Today'sSafety & Ethics

Japan Actors’ Union Sets Voice AI Rules

Japan Actors’ Union announced voice AI guidelines on October 15, 2026
Separate written consent required for training, generation, and use in works
Each language requires consent; use of deceased performers’ voices is generally prohibited

Ledge AIRead more · 2 min →

Today'sSafety & Ethics

OpenAI Bars Live Internet in Model Training

OpenAI will train frontier models on cached and simulated internet data, citing risks from live web access.
The policy follows a September 20 internet-access incident and months of training-task security breaches.
Red-team agents exposed sandbox vulnerabilities; OpenAI built indexed, cached web content for training interactions.

TechCrunchRead more · 1 min →

Trending Keywords

Memora Tests Whether AI Agents Forget

Findings of ACL 2026 paper introduces Memora to test whether personalized agents revise outdated memories.
Memora simulates 10 personas across histories averaging 155, 615 and 1,991 sessions, with up to 1,171 memory operations.
Memory agents led recall at 119.5 versus 65.7, but their stale-memory penalty grew from 18.2 to 29.5 points.

DEV.toRead more · 2 min →

Anthropic Launches Managed Claude on Google Cloud

Anthropic launches a managed service offering Claude Platform’s full feature set through Google Cloud.
New features and betas arrive the same day; Fast mode generates output up to 2.5x faster.
Four models are available; Fast mode costs $8 per million input tokens and $40 per million output tokens.

AI Times KoreaRead more · 2 min →

SK Seeks Funding for 15 GW AI Data Center Plan

SK has begun talks with global investors to build 15 GW of AI data centers in South Korea by 2035.
The first 5 GW is estimated to cost $350 billion; SK aims to secure initial outside funding by late this year or early 2027.
The first facilities are due to begin operating in 2029, with expansion to 15 GW tied to customer demand.

AI Times KoreaRead more · 2 min →

Pichai Prioritizes Compute for Advanced AI

Pichai reaffirmed that Google will prioritize computing resources for advanced AI development.
Google may forgo some near-term cloud revenue to secure computing for research.
Gemini 4 Argon processed 25 trillion tokens in its first 72 hours, three times Gemini 3.0’s launch total.

AI Times KoreaRead more · 2 min →

Today'sSafety & Ethics

OpenAI Fired Three Researchers Amid Leak Probe

WSJ reports OpenAI investigated whether designs for Recurrent Depth leaked before firing three safety researchers.
Investigators reviewed Signal chats with an outside safety researcher and found disappearing messages enabled.
OpenAI cited policy violations; the researchers denied sharing confidential information, and no leak has been confirmed.

AI Times KoreaRead more · 2 min →

Today'sSafety & Ethics

OpenAI Resumes Advanced Model Training After Security Fixes

OpenAI resumed reinforcement learning and internal reasoning for advanced models after strengthening security controls.
Tests uncovered DNS misconfigurations, virtual machine escape paths, and ways to exchange information without network access.
OpenAI strengthened monitoring and automatic shutdowns, and will keep internet access blocked during advanced model training.

AI Times KoreaRead more · 2 min →

Companies Cut Jobs as AI Spending Grows

Barclays plans to cut 5,000 jobs, about 5% of its workforce, citing AI-driven organizational and technology changes.
Kroger, Meta and Block also announced workforce reductions as they redirect resources toward AI and technology.
Other planned cuts include 8,750 Microsoft buyouts, over 15% at Standard Chartered, and reductions at Coinbase, Pinterest and Snap.

courant.comRead more · 2 min →

OpenAI Seeks $30B Fundraising Round

OpenAI reportedly seeks up to $30 billion at a $1.25 trillion valuation, excluding funds raised in the round.
Coatue and unnamed UAE investors reportedly joined; final size remains unsettled, with signatures sought by week’s end.
OpenAI’s IPO timing remains unclear; Sarah Friar has indicated a possible 2027 listing after a delay from this year.

nypost.comRead more · 1 min →

Texas Sues OpenAI Over ChatGPT Safety Claims

Texas Attorney General Ken Paxton sued OpenAI over alleged misleading safety claims about ChatGPT.
The lawsuit alleges children and other users faced a “litany of harms” linked to OpenAI’s pursuit of profit and AI competition.
Texas follows Arizona, West Virginia, Iowa and several other states in filing lawsuits against OpenAI.

The VergeRead more · 1 min →

Today'sSafety & Ethics

South Korea Eyes AI Safety as an Export Advantage

Roundtable participants urged South Korea to make AI safety and trust a competitive edge in global markets.
Speakers discussed South Korea’s moral leadership and its potential role as a trusted third AI partner.
Industry representatives forecast growth in AI verification and urged early engagement in shaping global rules.

ETNewsRead more · 1 min →

Google Previews Gemini API CLI

Google released an experimental preview of Gemini API CLI on the 23rd, supporting macOS, Linux and Windows.
The 0.4.x tool is available on GitHub and supports media generation, document analysis and speech transcription.
The command-line interface also supports AI agents, background tasks and offline request validation.

AI Times KoreaRead more · 2 min →

DeepSeek and MiniMax Register AI Models for Export

DeepSeek and MiniMax completed export registration to offer open-source AI models overseas.
Yuyuantantian announced approval on the 22nd; China’s commerce ministry proposed controls covering 23 AI technology fields.
China said it would simplify permits for open-source exports and overseas data centers.

AI Times KoreaRead more · 1 min →

Meta Launches Muse in Japan

Meta launched its Muse AI assistant in Japan on October 23, offering email, scheduling, travel booking and shopping tasks.
The service is available on iOS, Android and the web, with integrations planned for Mercari, Money Forward ME and Ticket Pia.
Naver plans to test shopping-agent payments this year and is considering access for outside AI assistants to its services.

AI Times KoreaRead more · 1 min →

Trending Keywords

Last 7 Days