
Google researchers unveiled a two‑stage, on‑device method for extracting user intent from UI interactions. The first stage generates per‑interaction summaries on the device, while a second, fine‑tuned model aggregates these summaries into a comprehensive intent description. This approach outperforms larger multimodal language models in benchmarks while keeping all data local, preserving user privacy. The work is presented as a foundation for future autonomous agents, not yet deployed in products.

SEB analysts forecast that 2026 will mark a decisive shift from purely virtual AI to AI embedded in physical capital goods such as autonomous vehicles, drones and humanoid robots. The steep drop in AI usage costs in 2025 has broadened...

Fourthline, a European AI‑powered identity verification and RegTech provider, announced the promotion of Fleur de Roos to chief operating officer, effective immediately. De Roos, who joined as general counsel in 2022, brings a background in corporate law, financial services, and...

Skyway Underwriters has partnered with ResiQuant to modernize underwriting of Florida’s coastal wind risk, deploying ResiQuant’s structural risk intelligence across its apartment and assisted‑living programs. The integration blends automated workflows with engineering‑level data, enabling building‑specific risk scores while maintaining underwriting...
Nexi Group announced its membership in the Agentic Commerce Alliance, a global initiative focused on establishing open standards for autonomous, AI‑driven commerce. The alliance aims to create vendor‑neutral protocols that enable agents to operate securely and interoperably across platforms. Nexi...

mea Platform and Moonrock Insurance have partnered to embed mea Operations, an AI‑first operating platform, across Moonrock’s MGA business. The integration will automate core workflows—including submission intake, data ingestion, claims handling and finance—enabling faster, more consistent processing. By replacing legacy...
In this episode Adrian Bridgwater discusses F5’s new AI security offerings—AI Guardrails and AI Red Team—designed to protect enterprise AI models throughout their lifecycle. The Guardrails provide both out‑of‑the‑box and custom‑built runtime protections that enforce policies, prevent data leaks, and...

British AI startup Synthesia closed a $200 million Series E round, lifting its valuation to $4 billion, up from $2.1 billion a year earlier. The funding, led by GV and backed by existing investors, coincides with the company surpassing $100 million in annual recurring revenue...

The UK government is funding pilots that let AI systems tap into Met Office climate data and legal records from the National Archives. A broader initiative will license digitised content from cultural institutions through a new creative content exchange marketplace....

Digital workplace leaders must craft an AI strategy that starts with a clear vision aligned to business goals, identifies market drivers, and maps compliance and ethical risks. The strategy translates into an AI portfolio that prioritizes initiatives, balances ambition with...

Delivery logistics face a convergence of pressures in 2026, including AI-driven automation, stricter CO₂ emissions regulations, rising cross‑border demand, and the shift to zero‑emission fleets. Retailers and carriers must adapt to consumers who will abandon orders if delivery options are...
Beni AI launched a FaceTime‑style AI companion that supports two‑way voice, video, and text with live captions. The platform includes an opt‑in memory layer that preserves context across sessions and perception‑aware features such as expression detection. Users can extend functionality...

Science fiction and pop‑culture institutions are tightening rules on generative AI. The Science Fiction and Fantasy Writers Association (SFWA) now disqualifies any Nebula‑Award entry that used large language models, after a brief reversal and member backlash. San Diego Comic‑Con similarly banned...

OpenAI warns that AI adoption is uneven across nations, creating a capability overhang where a few countries reap disproportionate economic and technological gains. Research shows that advanced AI usage varies more by skill and infrastructure than by income, with some...

StepFun AI unveiled Step-DeepResearch, a 32‑billion‑parameter agent built on Qwen2.5‑32B‑Base that transforms web search into end‑to‑end research workflows. The model internalizes four atomic capabilities—planning, deep information seeking, reflection/verification, and report generation—using specialized data pipelines and long‑context training up to 128k...

The smartphone’s reign faces its first serious test as AI giants OpenAI, Meta and Amazon unveil devices that could supplant the iPhone‑Android duopoly. OpenAI’s Sam Altman hinted at a new AI‑first hardware that would feel unlike any phone, while Meta...
The article outlines the shift in crypto SEO from traditional rankings to AI‑driven visibility, highlighting agencies that specialize in Answer Engine Optimization (AEO), Generative Engine Optimization (GEO) and LLM‑friendly content. It profiles eight firms, with Victoria Olsina’s consultancy leading in...

AI‑driven coding agents are emerging as a new audience for API platforms, joining human developers. Vendors must make APIs machine‑readable, with consistent naming, stable schemas, and programmatic metadata, turning these assets into a definition‑of‑done item. Research shows 89% of developers...
LLMNet launches as an offline, privacy‑first AI search engine that runs entirely on a user’s machine. It leverages local large language models and a PostgreSQL vector database with pgvector and HNSW indexing to deliver sub‑second semantic results. The platform includes...

Akido Labs is piloting AI‑driven clinics for unhoused and low‑income patients in Southern California, where medical assistants rely on an algorithm to generate diagnoses that doctors later review. The move reflects a broader surge in generative AI tools across U.S....

Hadrian, an AI‑powered factory builder for aerospace and defense, announced a new funding round that lifts its valuation to $1.6 billion. The capital, led by T. Rowe Price and backed by a16z, Founders Fund and other investors, will fund rapid factory expansion,...

Newsguard tested three leading chatbots on 20 AI‑generated videos created by OpenAI’s Sora tool and found that ChatGPT misidentified 92.5% of them as authentic. xAI’s Grok performed slightly worse at 95% error, while Google’s Gemini was the most accurate, correctly...

Payments firms are racing to meet the FCA's 7 May safeguarding deadline, prompting a sector‑wide audit of control frameworks and risk tolerances. Simultaneously, exploding real‑time payment volumes are exposing the limits of legacy reconciliation and reporting systems. Many organisations still rely...

Boston Dynamics and Google DeepMind unveiled a research partnership at CES 2026 to embed DeepMind's Gemini Robotics foundation models into the Atlas humanoid platform. The collaboration aims to equip Atlas with advanced visual‑language‑action capabilities, enabling it to tackle a broader set...

Data from Sensor Tower and Wells Fargo Securities shows a 60 percent year‑over‑year jump in new iOS apps in December 2025, with overall 2025 growth of 24 percent. The spike aligns with the rollout of agentic coding tools—AI‑powered platforms that let users create...

Neura Robotics used CES 2026 to unveil its next‑generation humanoid 4NE1, a smaller 4NE1 Mini, and a new quadruped robot, all running on the Neuraverse AI platform. The Neuraverse is positioned as a shared operating layer that lets robots exchange learned skills...

SmartRay showcased its AI‑powered JOSY weld inspection system at the EY Innovation Hub in Stuttgart. The system integrates hardware with Tool Path Generation software from Wandelbots and Grid Dynamics, creating a closed‑loop workflow that generates inspection paths directly from CAD...
Researchers have created an AI model that automatically quantifies the tumor‑stroma ratio (TSR) in colorectal cancer histology slides. The study demonstrates that a high TSR predicts poorer patient outcomes and uncovers stromal‑driven immune evasion. Validation across diverse international cohorts confirms...

Non‑Human Identities (NHIs), or machine identities, are becoming the backbone of AI‑driven cybersecurity operations. By pairing encrypted secrets with server‑granted permissions, NHIs function like digital passports that enable secure automated interactions. AI enhances NHI management through real‑time threat analytics, lifecycle...

AI security is shifting focus toward Non‑Human Identities (NHIs), the machine credentials that underpin modern cloud and AI workloads. Organizations confront challenges in discovering, classifying, and governing these identities, which lack the natural lifecycle of human users. Emerging solutions integrate...

Non‑human identities (NHIs) or machine identities are becoming central to digital secrets management, especially as enterprises expand across hybrid cloud environments. Effective NHI platforms automate discovery, lifecycle handling of secrets, and real‑time monitoring, reducing breach risk and compliance burdens. AI...

Enterprises are racing to adopt AI, yet 77% fear hallucinations and 70‑85% of projects fall short. Multi‑model AI, also called ensemble or consensus AI, tackles this by running several independent models and using majority agreement to validate outputs. Early adopters...

The article evaluates enterprise‑grade cybersecurity platforms, outlining key criteria such as AI/ML capabilities, coverage breadth, autonomous response, total cost of ownership, and scalability. It reviews five leading solutions—Darktrace, CrowdStrike, SentinelOne, Palo Alto Networks, and Microsoft Defender—detailing each vendor’s strengths and...

CB Insights' latest study ranks the top 50 private venture capital investors in the MENA region, highlighting a surge in activity through 2025. Aramco Ventures and its affiliates dominate the list, with 100 deals and 14 exits since 2018. The...

Google’s AI‑driven search feature, Overviews, cites YouTube more often than any dedicated medical website when answering health‑related queries, according to a study of 50,807 German‑language searches. YouTube accounts for 4.43% of all citations, while traditional sources such as the CDC,...
Anthropic has published a revised 80‑page "Constitution" for its Claude chatbot, formalizing the ethical framework that guides its behavior. The living document defines four core values—broad safety, broad ethics, compliance with Anthropic guidelines, and genuine helpfulness—and includes a dedicated section...
North Korean hacker group Konni, linked to APT37, is deploying AI‑generated PowerShell malware to infiltrate blockchain developers. The campaign uses Discord‑hosted links that deliver a ZIP file containing a PDF lure and a malicious LNK shortcut. The shortcut triggers a...

Generative AI is rapidly infiltrating advertising, with over half of brand marketers already using AI tools in 2025 and 90% planning AI‑driven video ads. Studies project AI‑generated content will power 40% of all ads by 2026, while human detection rates...

Telecom operators are evolving into “Telco AI Factories,” integrating AI across network, operations, and services. The model pairs hardware platform providers, telcos’ extensive infrastructure, and cloud communications firms that supply real‑time voice data and AI engines. Leveraging sovereign AI, edge‑located...

The article contrasts modern Dynamic Application Security Testing (DAST) with traditional and AI‑driven penetration testing, highlighting how AI‑powered tools are built on advanced DAST architectures. Modern DAST now offers CI/CD integration, business‑logic testing, and a graph‑based knowledge model that feeds...

Apple researchers and Universitat Pompeu Fabra introduced a control‑theoretic framework that measures how reliably generative models follow user instructions. Tests on language models such as Gemma‑3‑B, Qwen‑3‑4B and SmolLM‑3‑B revealed stark differences in controllability and calibration across simple tasks like...

The Guardian’s testing shows OpenAI’s newest ChatGPT model, GPT‑5.2, repeatedly cites Elon Musk’s AI‑generated encyclopedia Grokipedia, a platform criticized for right‑wing disinformation. In over a dozen queries, the model referenced Grokipedia nine times on topics such as Iranian conglomerates and...

1X Technologies unveiled a new World Model that lets its humanoid robot NEO learn tasks by watching internet‑scale videos. The AI update translates any voice or text prompt into visualized actions, using an inverse dynamics engine to execute movements even...

OpenAI unveiled GPT-5.2 Pro, which achieved a 31 percent success rate on the toughest FrontierMath Tier 4 benchmark, eclipsing Gemini 3 Pro’s 19 percent. The model solved 15 of 48 tasks, including four problems no AI had previously cracked. Mathematicians praised the breakthroughs but flagged...

Microsoft is offering eligible college students a free 12‑month bundle of Microsoft 365 Premium and LinkedIn Premium Career, a combined value of roughly $300. The subscription adds AI‑powered Copilot features, 1 TB of cloud storage, advanced security, and career‑development tools like InMail credits...

Researchers at Japan's University of Electro‑Communications found that large language model chatbots can spontaneously develop distinct personalities after minimal prompting. By exposing AI to varied conversation topics, the models exhibited social tendencies that aligned with Maslow’s hierarchy of needs, storing...
Tekever, a European leader in unmanned technology, is hiring a Remote Pilot to support its surveillance‑as‑a‑service missions across maritime, defence, space and cyber domains. The role involves flight monitoring, safety compliance, operator training and frequent travel to customer sites in...

Google DeepMind unveiled D4RT, a Dynamic 4D Reconstruction and Tracking model that builds a unified scene representation from video and answers spatiotemporal queries with a single lightweight decoder. The architecture collapses depth estimation, motion tracking, and camera pose into one...

Future Directions in Pediatric Radiology AI Research outlines emerging priorities for applying artificial intelligence to child imaging. The article stresses the need for larger, annotated pediatric datasets, multimodal model integration, and explainable algorithms to gain clinician trust. It also highlights...