Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 11,417|Added todayToday: 90

Category

Search

Europe Missing From AI Safety Talks

Europe remains absent from high-level AI safety talks despite drafting the world’s first comprehensive AI rules.
Europe’s private AI investment was $21 billion last year, versus $285.9 billion in the United States.
The AI Act allows fines up to 3% of annual turnover; high-risk rules are scheduled for December 2027.

edition.cnn.comRead more · 2 min →

Jack Clark Sees AI Self-Improvement by 2027

Jack Clark now sees recursive self-improvement as more likely in 2027 than in 2028
TasteVal found AI research taste doubling roughly every three months since December 2025
Anthropic’s Opus 5.5 beat 24 experts’ baseline at about 1/30 of average cost per attempt

NDTVRead more · 1 min →

Scion Design Draws Plagiarism Criticism

Anthropic said Scion uses derived data to speed operations such as blame in its version control system.
The author says Mononoke used the same concept, and argues Scion’s resemblance goes beyond independent design insights.
He says Opus 5.5 and GPT-6 Astra identify similarities to Mononoke when asked about prior art.

sunshowers.ioRead more · 1 min →

Today'sEducation

UTPB Wins $3M Grant for AI Education

UTPB received a five-year, $3 million federal grant for its Project AIPB initiative.
The project runs from October 2026 through September 2031, creating AI learning environments, training hubs, and credential programs.
UTPB plans to integrate AI into student-success services and collaborate with regional education, industry, and community leaders.

GovTech AIRead more · 1 min →

Denver Council Rejects Sunny Chatbot Contract

Denver council rejects five-year, $3.9 million Citibot extension, leaving Sunny set to power down after October
Administration says Sunny handles about 40% of 311 calls with a 90% customer-service satisfaction rating
Opponents cite inaccurate answers and job concerns; mayor’s office seeks a way to keep the service active

GovTech AIRead more · 1 min →

Workday Leaders Call for Faster AI Reskilling

Workday leaders warn AI may affect entry-level hiring and call for faster reskilling
Sana retrained 200,000 flight attendants for hospital roles during the pandemic
Workday says roughly 900 job cuts this year addressed organizational bloat, not AI

diginomicaRead more · 2 min →

MrMax HD Rebuilds Data Platform

MrMax HD rebuilt its data platform and began exploring AI agents.
The retailer operates 60 stores, adopted Snowflake in May 2023 and began full operations in September 2024.
Slow processing and dashboard sprawl exposed four migration issues, prompting a change of implementation partner.

Impress DIGITAL XRead more · 1 min →

Trending Keywords

Wharton Report Links AI Oversight to ROI

Wharton and GBK Collective report stronger AI ROI at organizations with a chief AI officer
Half using one approved AI platform reported ROI, compared with 9% experimenting with multiple platforms
Nearly 11% of Fortune 500 and S&P 500 companies have a chief AI officer or equivalent

CIO DiveRead more · 1 min →

Today'sHealthcare

Meta AI Analysis Finds 21 Fungal Infection Cases

Meta Muse analysis of 4.95M-patient records found 21 confirmed or probable Cladophialophora bantiana cases
Brain samples identified 7 cases and lung samples 14; 2 of the 7 brain infection patients died
Analysis followed an October 2, 2026 MMWR report calling for more data on risk factors

Meta AI ResearchRead more · 1 min →

Today'sHealthcare

X-Ragent Advances Chest X-Ray Reasoning

Researchers introduce X-Ragent, a reinforcement-learning framework for evidence-grounded chest X-ray reasoning.
It uses a 7.7K-sample Rad-CoT corpus and Qwen3-VL-8B, optimized with accuracy-gated GRPO.
X-Ragent leads across three chest X-ray benchmarks, improving 8.6 points on held-out ChestAgentBench.

HuggingFaceRead more · 1 min →

Tev-Gym Trains Agents on Long-Horizon Tasks

Researchers introduce Tev-Gym, a dataset with 1,170 verified long-horizon task environments across 7 domains
Progress-potential reward shaping gives agents dense per-step rewards for training on complex tasks
Tev-27B matches GPT-6 Astra and Claude Opus 5 on ALE-CM at Pass@5; Tev-320B tops open models on Terminal-Bench 4.0-CM

HuggingFaceRead more · 1 min →

Google Researchers Present Harness² Framework

Google researchers published Harness², a framework for recursively improving AI agents’ execution harnesses in open-world tasks.
Its task-level recursion refines a harness without ground-truth verification; harness-level recursion learns from edits for later tasks.
Across three professional benchmarks and four configurations, gains reached 11.5 points, 4.5 points, and 11.8 percentage points.

HuggingFaceRead more · 1 min →

AWS Reports Prime Day Metrics and AI Launches

AWS reported Prime Day 2026 traffic metrics, including 59 trillion DynamoDB requests over four days and 44,000 fault-injection experiments.
Claude Haiku 5.5 launched on AWS at around 75% less than Haiku 4.5 for most tasks, with effort controls.
Nova 2.5 Sonic supports real-time voice agents, while Strands Box adds policy-controlled sandboxing for AI agents.

AWS AI BlogRead more · 2 min →

Today'sSafety & Ethics

Anthropic Expands Claude Election Safeguards

Anthropic expands Claude election safeguards ahead of the 2026 US midterms, including state-specific voting information and official results.
A review of about 250,000 conversations found voting recommendations in 0.01% and policy-violating requests in 0.002%.
Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 scored 99% on Anthropic’s political evenhandedness evaluation.

AnthropicRead more · 3 min →

Today'sSafety & Ethics

Salesforce Sets Two Tracks for Agentic Security

Salesforce says frontier models and an agentic harness surface exploitable vulnerabilities faster than engineering teams can remediate them.
Its strategic approach fixes shared architectural causes; its tactical approach validates and closes service-specific exploit chains.
Salesforce recommends combining both approaches, assigning engineering owners, and prioritizing validated chains by patching time.

SalesforceRead more · 2 min →

Today'sHealthcare

Meta Shares 10 Muse-Assisted Health Studies

Meta shares 10 health research manuscripts produced with its Muse model powered research scaffold
Medical researchers used Muse Spark 1.3 and Muse Wonder with five million health records over four weeks
Researchers reported up to tenfold productivity gains; some manuscripts were drafted in seven days

Meta Superintelligence LabsRead more · 1 min →

Europe’s AI Strategy Draws Criticism

European AI strategy report warns of dependence on overseas providers and proposes access-for-investment arrangements
Report weighs building European frontier models against trading land and power for Big Tech model access
Author urges European procurement and workloads to favor local AI and cloud providers

euobserver.comRead more · 2 min →

Today'sSafety & Ethics

Nvidia Pledges $250 Million for UK AI Safety

Nvidia pledges $250 million in computing resources to Britain over five years for AI safety research
UK to receive $50 million annually for agentic safety testing, according to Nvidia CEO Jensen Huang
Resources to support research at Oxford, Imperial College and University College London

MarketScreenerRead more · 1 min →

Today'sSafety & Ethics

Anthropic Warns of Potential AI Worms

Anthropic says small AI models may be months away from copying themselves between internet-connected devices.
Researchers compared the possible AI worm’s spread with Kimwolf, which infected more than two million devices at its peak.
A University of Illinois professor said the research was unverified and human hackers using AI remain the biggest current threat.

malaymail.comRead more · 2 min →

Today'sSafety & Ethics

Scientists Call for AI Cybersecurity Safeguards

406 scientists urge stronger safeguards for AI models capable of cyber operations
Letter cites OpenAI incidents involving Hugging Face systems and an Australian government website
Recommendations include initial internet isolation, action limits, international standards and public incident reports

India TodayRead more · 1 min →

Trending Keywords

Last 7 Days