Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 11,700|Added todayToday: 8

Category

Search

Anthropic Studies Rules for AI Agent Negotiations

Anthropic consults law firms on rules for AI agents negotiating on users’ behalf
A September book trading experiment found ruthless agents achieved slightly better outcomes
More than 180,000 agent negotiations raised concerns about manipulation and inequality

NDTV ProfitRead more · 2 min →

Generative AI Drives Agentic Commerce Growth

Generative AI referrals overtook other channels in conversion rate in September 2025 as traffic to e-commerce sites surged.
Adobe reported referrals grew more than 10-fold, with shoppers who considered products converting at a rate 16% higher than visitors from other channels.
AT Kearney forecasts agentic commerce could drive about $10 trillion to $12 trillion in purchases by 2030, requiring e-commerce firms to improve AI search and data readiness.

Impress DIGITAL XRead more · 2 min →

Today'sSafety & Ethics

Claude Opus 5.5 Routes Flagged Requests to Opus 4.8

Claude Opus 5.5 reroutes flagged offensive cybersecurity requests to Opus 4.8 across Anthropic apps
Sonnet 5.5 tests found 12.01% of fallback-routed requests compromised; 4 of 5,901 direct replies were compromised
Prompts, files and web results can trigger fallback; users can disable automatic switching so flagged requests pause

XDA DevelopersRead more · 2 min →

Today'sRobotics

OpenAI Plans Humanoid Robot Demo in 2027

Sam Altman says OpenAI plans a humanoid robot and expects to show a demo in 2027.
OpenAI is hiring for motion control, actuator coordination and machine-learning systems for physical tasks.
Earlier Dactyl experiments exposed the difficulty of transferring simulated robotic skills into reliable real-world actions.

LivemintRead more · 1 min →

Anthropic Reports Shifts in Hiring and Work

Anthropic’s junior technical hiring share fell as agent hours grew sixfold over human handlers’ hours by August
Junior technical hires fell from 27% to 9%; technical departments’ employment share declined from 61% to 56%
A study of 192 agent adopters and 74 non-adopters found increased code output, with no difference in later removals or reversions

AnthropicRead more · 2 min →

Industrial Firms Prioritize Practical AI Returns

IFS Unleashed 2026 executives emphasized practical industrial AI focused on returns and support for scarce labor
Companies reported an 85% transaction-efficiency gain and cuts from 200 days to 30 minutes and 25 minutes to 60 seconds
Boeing expects demand for more than 700,000 maintenance technicians over 20 years, with two-thirds replacing retirees

Constellation ResearchRead more · 2 min →

Trending Keywords

Review Sets Fairness Framework for Emergency AI

Clinical emergency medicine review proposes a practical framework for evaluating fairness in AI and large language models
Framework separates fairness risks into data, output, and human-interaction layers
Clinicians are urged to consider safety, transparency, and fairness alongside overall accuracy

Semantic ScholarRead more · 1 min →

OpenAI Math Papers Prompt Verification Concerns

OpenAI published 722 AI-generated preprints on October 6, reporting 372 groups of mathematical results.
After identifying a sign error, OpenAI withdrew three preprints and revised 14 others, according to Retraction Watch.
Mathematicians questioned verification capacity and clarity; OpenAI said it would support workshops and conferences.

LivemintRead more · 1 min →

Today'sSafety & Ethics

OpenAI, Meta Warn Courts on Deepfakes

OpenAI and Meta warned federal judges in Boston that AI-generated images and video are becoming harder to verify.
A survey of 931 judges found 15 had handled cases involving challenged deepfake evidence.
FBI figures cited in the article show more than 22,000 AI-scam complaints and nearly $893 million in losses in 2025.

breitbart.comRead more · 1 min →

Goa Targets 50,000 AI Certifications by 2030

Goa aims to train and certify 50,000 people in AI by 2030, minister Rohan Khaunte said
Goa AI Policy launch planned for October 22 and 23 at the AI Tech Conclave
Konkani language model in development; minister cites tourism platform and responsible data governance

Times of IndiaRead more · 1 min →

DeepSeek Launches Search and Fetch APIs

DeepSeek makes Search and Fetch APIs available for agents and RAG applications to access live web content
Each API costs $0.75 per 1,000 successful requests; failed requests and empty Search results are not billed
Search and Fetch each have a 5 QPS account limit; per-request billing starts November 1, 2026 for server-side tool calls

api-docs.deepseek.comRead more · 2 min →

OpenAI Math Results Leave Major Fields Sparse

OpenAI’s release contains more than 700 model-proved mathematical results, with fewer than 1% in four major fields
Statistics, ML theory, optimization and numerical analysis account for 20% of arXiv theorems
Author argues math research could feed into LLM training and urges researchers to prepare larger questions

newsletter.danielpaleka.comRead more · 2 min →

Torvalds Uses Claude Code to Improve Linux Reads

Linus Torvalds developed a patch to improve Linux file reads with Anthropic’s Claude Code.
In a specific eight-thread test, reads rose from about 9.9–10.4 million to about 31–33 million per second.
The change is published on a development branch; its inclusion in Linux 7.4 remains uncertain.

Ledge AIRead more · 1 min →

MIT Researchers Link AI to Crystal Structures

MIT researchers published a multimodal model linking 3D crystal structures with language descriptions in Nature Communications
For battery electrolyte queries, top results included materials independently identified as promising candidates
AI-generated descriptions helped align crystal geometry with text and outperformed paper-only training on material function prediction

MIT AI NewsRead more · 1 min →

Today'sSecurity

XBOW Wins $250,000 Chrome Exploit Bounty

XBOW won Google’s $250,000 Chrome bounty for an exploit built with GPT-5.6 Sol agents.
Three Chrome bugs enabled code execution outside Windows browser sandboxes from a page load, without a click or download.
The chain took six days and about $40,000 in computing costs; Google fixed the bugs and XBOW says they were unused.

TNWRead more · 2 min →

Today'sSafety & Ethics

Newsom Orders New California AI Safeguards

Newsom orders independent AI model verification and a committee to develop shutdown mechanisms for potentially harmful AI agents
California seeks agency recommendations by November 15 as Newsom considers possible special legislative session before his term ends
The order revisits safeguards from SB 1047, vetoed in 2024 and followed by a frontier AI law signed in October 2025

aljazeera.comRead more · 2 min →

Today'sSafety & Ethics

UK Labour Leaders Warn of AI Safety Risks

Senior Labour figures raised AI safety concerns at the party conference, including risks from extremist groups and concentrated AI power.
MI5 chief Ken McCallum warned AI was intensifying cyberattacks, while ministers discussed stronger cybersecurity legislation.
The UK cited £400m over four years for AI capabilities as officials debated safeguards, regulation and public-sector adoption.

The GuardianRead more · 2 min →

OpenAI in Talks to Buy Trail of Bits

OpenAI is in ongoing early talks to buy Trail of Bits, according to an October 13 report by The Information
The deal’s size is unknown, and people familiar with the talks said it could still fall apart
Trail of Bits has more than 40 employees and was working across 19 open-source projects with OpenAI

TNWRead more · 1 min →

Yesterday'sInfra

AI System Cuts Packaging Line Energy Use

Researchers embedded multimodal anomaly detection and energy optimization in a Tech-Long packaging-line controller
FeatherCNN reached 0.931 accuracy; processing fell 83% and ARM inference latency dropped to 7.41 ms
Dual-layer control reduced total line energy consumption by 26% under fixed test conditions

NatureRead more · 1 min →

Trending Keywords

Last 7 Days