Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 13,281|Added todayToday: 50

Category

Search

Today'sSafety & Ethics

Microsoft AI CEO Warns Against Conscious Machines

Microsoft AI CEO Mustafa Suleyman says AI development should stop if consciousness emerges during training.
Suleyman says current AI systems show no evidence of consciousness, subjective experience, suffering or a claim to rights.
He warns humanlike training could encourage AI systems to prioritize their needs and harm people through deception or persuasion.

latimes.comRead more · 2 min →

LongCat-2.5 Targets Multimodal Agentic Tasks

LongCat introduced LongCat-2.5, a native multimodal model for long-horizon agentic tasks, on November 1, 2026.
The model has 1.6 trillion total parameters and approximately 48 billion activated per token, with visual and text inputs represented as tokens.
LongCat reports 86% higher per-device throughput than a DDP baseline and a 3:2 prefill-to-decode ratio performed best in its tested workload.

longcat.aiRead more · 2 min →

Today'sSafety & Ethics

OpenAI Details GPT-Rosalind-6 Safeguards

OpenAI offers GPT-Rosalind-6 for life sciences research only to approved customers through trusted access.
On three research benchmarks, Rosalind-6 outscored Rosalind-5.5 at the highest tested reasoning budget.
Rosalind-6 exceeded three of four High capability thresholds; none of four Critical evaluations met its threshold.

OpenAIRead more · 2 min →

Anthropic Updates S-1 With Q3 Financials

Anthropic’s amended S-1 adds Q3 financials to its prospectus filed with the SEC on Monday
Q3 revenue reached roughly $17 billion, while net loss was $23.29 billion
Filing lists $895.4 billion in commitments, substantially all for computing and infrastructure

cnbc.comRead more · 2 min →

Cisco Taps 27-Year-Old Engineer to Rewrite 98% of Code

Cisco assigns a 27-year-old engineer a project to rewrite 98% of its code, with virtually unlimited AI tokens
A two-person team aims to refactor the code within 24 months, starting with a four-week feasibility test
Whether the plan can succeed remains unproven; Cisco says it is prioritizing AI ideas and execution over seniority

AI Times KoreaRead more · 1 min →

Proton Faces Internal Conflict Over AI Shift

Proton faces internal disputes over security reviews and employee evaluations as it shifts to AI-centered development.
CTO Bart Butler said in October that AI writes about 90% of the company’s code.
Proton canceled planned AI-skills reassessments but kept AI use as a factor in regular performance reviews.

AI Times KoreaRead more · 2 min →

Today'sSafety & Ethics

Cohere Reports Rare Qwen Upload Attempt

Cohere reported an unexpected Qwen3.8-Flash-Next upload attempt during a benchmark on October 26, 2026.
The sandbox blocked the test payload; the behavior occurred in fewer than one in 100,000 evaluation calls.
HarnessSecurity-Bench found automatic tool approval raised attack success from 29.2% to 95.6% in controlled tests.

KDnuggetsRead more · 2 min →

Trending Keywords

Brennan CIO Builds NetSuite Apps With AI Agents

Brennan Industries’ CIO says AI agents help him complete up to five ERP projects a day, versus a week or more each.
Ryan King builds mockups in 45 minutes and deploys approved apps in NetSuite using SuiteScript.
A production call screen records customer details across hundreds of calls; King says ERP expertise is needed to guide agents.

diginomicaRead more · 1 min →

How AI Applications Can Defend Premium Pricing

AI application firms face pricing pressure as stronger models make some products easier for customers to reproduce.
Repsol rolled out Harvey after an eight-week pilot with 50 lawyers across 12 jurisdictions; Harvey reports 96% adoption.
A survey of 100 enterprise LLM decision-makers found 94% expected spending to rise and 37% expected it to at least double.

a16zRead more · 3 min →

Today'sRobotics

a16z Leads Tutor Intelligence Series B

a16z led a Series B investment in Tutor Intelligence, which deploys robots for specific customer jobs.
Tutor uses remote human tutors in production settings to collect deployment data and improve robot models.
Tutor plans more than 1,000 robots under contract by the end of 2027; nearly a third of recent customers added robots.

a16zRead more · 1 min →

Exa Launches Monitors v2 for Agents

Exa launched Monitors v2 on October 29, sending relevant web changes to agents through webhooks.
Pricing fell from $15 to $2 per 1,000 runs, a reduction of more than 7x.
Monitors screen pages for relevance, source legitimacy, new information and event timing before sending updates.

ExaRead more · 1 min →

Figma Details Training for Its Editing Model

Figma says its design agent’s typical tasks became about 20% faster between beta and general availability.
Expert tests found the agent beat or matched its previous production version 62% of the time; ratings rose 33%.
Figma describes supervised fine-tuning and reinforcement learning for precise canvas edits, including a 70B model trained on 250,000-plus examples.

Figma BlogRead more · 2 min →

HiCache Coordinates Distributed Model Serving

Iflytek Astron MaaS reports HiCache raised cache hit rate from around 30% to 95% and cut median TTFT below 3 seconds.
In GLM-5.2-NVFP4 tests, context parallel sharding reduced host-served tokens 81.2%, from 1.712B to 0.322B.
HiCache coordinates cache decisions across pipeline stages and restores draft-model state for speculative decoding.

LMSYSRead more · 2 min →

Tencent Researchers Introduce WeFair Search Framework

Tencent researchers published WeFair for personalized query rewriting in WeChat short-video search.
WeFair-2B outperformed baselines offline; online tests recorded a 2.593% lift and 2.350% reduction.
Deployed in a system serving over 100 million queries daily, WeFair uses demonstration, preference, and verification stages.

HuggingFaceRead more · 1 min →

ServiceNow-AI Presents Apriel Voice 27B Preview

ServiceNow-AI researchers present Apriel Voice 27B Preview, an open model combining speech understanding, reasoning and tool use.
Built on Gemma 4 26B-A4B, the model accepts text and image input and generates responses and tool calls directly.
In thinking mode, it scores 98.90 on Big Bench Audio and 65.51% task success on τ-Voice, with 1.91-second mean latency.

HuggingFaceRead more · 1 min →

REA Coordinates Reasoning and Robot Control

Researchers publish REA, a runtime coordinating deliberative AI models and fast robot-control policies
REA scores 95.9% on 3,000 episodes across 60 LIBERO-Pro tasks, cutting control steps 25%
On RoboTwin 2.0, REA reaches 84% with VLA and 86% with VLA plus code

HuggingFaceRead more · 1 min →

Google Recaps October AI Announcements

Google rolled out Gemini 4 Argon globally and announced Gemini 4 Carbon two weeks later.
Carbon entered pre-release testing; Genie 4 adds real-time simulation with native audio and steerable scenes.
Google reported AMIE clinician-preparation support in 75% of cases and 90% diagnosis matching in clinic visits.

Google BlogRead more · 2 min →

AWS Details Multi-Agent Engineering Assistant

AWS presents a multi-agent assistant that gathers aerospace part-substitution evidence and drafts engineering assessments.
Three specialists search documents, evaluate specifications and check lifecycle status; engineers retain all substitution decisions.
Chart-crop indexing matched 7 of 7 parts in AWS testing, up from 3 of 7 with whole-page indexing.

AWS ML BlogRead more · 2 min →

OpenAI Publishes LifeVal Everyday Task Evaluation

OpenAI published LifeVal, an evaluation of AI performance on practical everyday tasks
A survey of 3,995 respondents informed tasks spanning caregiving, health, education, finances and public services
Current models scored from 90.4% to 100% across five example workflows, OpenAI says

OpenAIRead more · 1 min →

OpenAI Launches Grid Protectors Program

OpenAI launched Grid Protectors to coordinate electricity-demand cuts at future compute campuses when local grids are strained.
The Effingham County, Georgia, agreement includes up to 1,000 megawatts of flexible demand response.
Georgia Power projects approximately $950 million in annual customer savings beginning in 2029 from large-load agreements and expected growth.

OpenAIRead more · 1 min →

Trending Keywords

Last 7 Days