This Week in NLP #356
Keep up with what happened in NLP in the week ending Friday 12th September 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Broadcom is reportedly helping OpenAI develop a custom AI chip using advanced packaging technologies to reduce reliance on Nvidia and lower model costs. [The Register]
Microsoft has partnered with Anthropic to integrate its AI models into Office 365 platforms, signaling a potential shift away from its close relationship with OpenAI. [ZDNet]
Mistral AI secured a €1.7bn funding round from ASML, doubling its valuation to €11.7bn and marking a victory for European tech sovereignty. [Tech.eu]
OpenAI has investigated why LLMs hallucinate, proposing evaluation methods that penalize confident errors and reward expressions of uncertainty. [TechCrunch]
Perplexity secured US$200m in funding, reaching a US$20 billion valuation while challenging Google's search dominance with AI-powered conversational answers. [TechCrunch]
Now read on for everything else that happened in NLP this week.
Making News:
What makes this newsletter different?
This Week’s Topics:
If you’re reading this in a desktop web browser, you can access a navigation menu at the left-hand edge of this window.
The Generative AI Wars
Black Forest Labs has signed a multi-year contract with Meta to provide AI image generation tools for a substantial financial commitment. [Sifted]
Google introduced AI Plus, a new affordable AI subscription tier targeting emerging markets like Indonesia, offering enhanced Gemini features and storage. [9to5Google]
Meta is reportedly recruiting US-based contractors to develop Hindi-language AI chatbots for Instagram, Messenger, and WhatsApp platforms. [DNA India]
Microsoft has bundled its Sales, Service, and Finance Copilots into the $30 Microsoft 365 Copilot package, reducing individual pricing and simplifying its AI offerings. [The Register]
And Microsoft has released a preview of Visual Studio 2026, offering deeper AI integration, a design refresh, and incremental improvements to its development environment. [The Register]
Nebius signed a US$17.4B multi-year AI infrastructure deal with Microsoft, beginning with a New Jersey data center project. [TechRadar]
OpenAI’s ChatGPT dominates the AI chatbot market with over 80% market share, while Microsoft Copilot and Perplexity compete for second place. [TechRadar]
OpenAI is reorganizing its Model Behavior team, merging it with the Post Training team and shifting focus to more nuanced AI model interactions and personality development. [TechCrunch]
OpenAI is considering relocating from California due to regulatory challenges surrounding its nonprofit-to-for-profit conversion, which could jeopardize its US$19B funding. [TechCrunch]
Oracle reportedly signed a massive US$300B cloud computing contract with OpenAI, spanning five years and beginning in 2027. [TechCrunch]
… but the Register notes OpenAI’s inability to afford the massive bill. [The Register]
OpenAI has dramatically increased its spending forecast to US$115B by 2029, driven by massive investments in computing infrastructure, model training, and talent acquisition. [The Decoder]
xAI has established a new engineering hub in Seattle, seeking talent with salaries ranging from $180,000 to $440,000 annually. [GeekWire]
AI Supremacy
Anthropic has restricted access to its AI models for companies majority-controlled by entities from China, Russia, Iran, and North Korea due to national security concerns. [The Decoder]
China’s AI Plus initiative aims to leverage AI as a strategic tool for economic transformation, addressing demographic challenges and boosting productivity across scientific, industrial, and governmental sectors. [Trivium]
Nvidia continues selling AI chips to Chinese tech giants like Alibaba and ByteDance despite Beijing’s regulatory pushback, with strong demand persisting due to domestic chip shortages. [Asia Financial]
And Nvidia has aggressively lobbied against congressional restrictions on selling AI chips to China, characterizing proposed export limits as driven by ‘doomer’ paranoia and threatening US technological leadership. [The New York Times]
Sovereign AI
A report by Antler reveals European tech startups, powered by AI and technical founders, are now scaling faster than ever, challenging previous misconceptions about European tech ecosystem limitations. [Tech.eu]
Jupiter, Europe’s first exascale supercomputer, launched at the Jülich Supercomputing Centre to advance scientific research, AI, and innovation across multiple domains. [ITPro]
Here’s a backgrounder on Mistral AI, the French AI startup that has rapidly grown to become a promising European competitor to OpenAI. [TechCrunch]
Pelagos Data Centres is planning a £1.8bn, 250MW datacentre in Gibraltar to support AI technologies and attract significant private investment by 2033. [Computer Weekly]
Qlik announced a new cloud region in Canada, powered by AWS, to support data sovereignty, lower latency, and enable governed AI for Canadian organizations. [Business Wire]
Second Front Systems will launch its Game Warden platform on AWS European Sovereign Cloud, enabling compliant and secure cloud deployment for European organizations. [Business Wire]
Sovereign Australia AI aims to develop an ethical, locally-trained LLM that compensates Australian creators while maintaining national technological sovereignty. [SmartCompany]
The UK government is establishing AI Growth Zones in de-industrialised areas to expand sovereign compute capacity, attract investment, and support its ambition to become an AI superpower by 2030. [Computer Weekly]
Feature Creeps
Anthropic’s Claude has introduced file creation capabilities, enabling users to generate Excel spreadsheets, documents, PowerPoint slides, and PDFs through conversational interactions with AI. [Anthropic]
Copilot has introduced four new File Explorer skills for Microsoft 365 users, enabling file summarization, comparison, and questioning directly through OneDrive. [ZDNet]
GitHub users are increasingly frustrated with Copilot’s intrusive AI features, demanding the ability to disable or block its code generation and review capabilities. [TechRadar]
Google has expanded its AI Mode search experience to five new languages, leveraging Gemini 2.5’s advanced capabilities to provide locally relevant and nuanced search results. [Google]
Google has launched Simplify, an AI-powered feature in its iOS app that rephrases complex text to enhance readability while preserving original meaning. [Google]
And Google’s Gemini has added audio file upload capabilities, enabling users to transcribe, summarize, and extract key information from 10-minute recordings. [TechRadar]
Grammarly expanded its AI writing assistance to Spanish, French, Portuguese, German, and Italian, offering grammar corrections, rewrites, and translations for multilingual users. [Business Wire]
Microsoft is redesigning the Copilot app with a Start menu-like interface, potentially hinting at future Windows 12 design changes. [TechRadar]
And Microsoft is preparing to expand AI collaboration features, enabling team-based work with AI and human partners across various roles and industries. [GeekWire]
OpenAI has added a conversational branching feature to ChatGPT, enabling users to explore multiple conversation paths without losing the original thread. [TechRadar]
OpenAI has extended its ChatGPT Projects feature to free users, enabling conversation organization with customizable instructions and file references. [Engadget]
Snapchat introduced an AI Imagine Lens that transforms selfies into creative images based on user-typed prompts, initially available to premium subscribers. [TechRadar]
Hype Bubble?
A survey by ABBYY revealed business leaders’ challenges with generative AI, including integration difficulties, governance issues, and the rise of shadow AI. [Business Wire]
Anthropic's CEO Dario Amodei’s prediction that AI would write 90 percent of code in six months proved wildly inaccurate, with research showing AI actually hindering software development. [Futurism]
Researchers at Arizona State University have debunked AI’s ‘reasoning’ claims, revealing that LLMs rely on pattern matching rather than genuine logical inference. [ZDNet]
Cohere co-founder Nick Frosst challenged the notion that scaling computing power will lead to AGI. [BetaKit]
Gartner predicts that by 2030, AI will transform IT work, with bots performing 25 percent of tasks and AI assisting humans in 75 percent of workloads. [Ars Technica]
And Gartner predicts that by 2028, no Fortune 500 companies will fully eliminate human customer service, emphasizing the irreplaceable role of human agents. [Business Wire]
Meanwhile, Microsoft expects 1.3 billion AI agents by 2028, but the technology remains largely unproven, with Gartner predicting 40% of agents will be abandoned by 2027 due to poor performance and limited productivity gains. [ITPro]
And Microsoft's UK government pilot of 365 Copilot found high user satisfaction but no clear evidence of productivity gains. [TechRepublic]
This piece argues that NANDA’s controversial MIT-affiliated report claiming 95% of enterprise AI fails lacks credible data, methodology, and substantiation, contradicting real-world evidence of successful AI implementations across multiple industries. [Futuriom]
Nvidia's CEO Jensen Huang believes AI and robotics will make humans busier, not less, while potentially improving overall life quality and productivity. [Gizmodo]
OpenAI and other AI companies are facing growing skepticism about their technology’s potential, with investors worried about overvaluation and real-world performance challenges that could trigger another AI winter. [Fortune]
A survey by the US Census Bureau revealed a decline in corporate AI adoption, with usage dropping from 14% to under 12% among larger companies, challenging the narrative of an AI revolution. [Futurism]
Big Iron
Cadence has added Nvidia's GB200 NVL72 Superpod to its digital twin platform, enabling datacenter operators to simulate complex AI infrastructure before massive investments. [The Register]
Dell's AI server revenues surged to $16.8bn in Q2, driven by AI demand, while storage sales lagged behind due to performance limitations and ongoing product development. [Computer Weekly]
Intelliflex has partnered with Supermicro to develop rapid-deploy, high-density AI data center solutions leveraging their complementary technologies. [EIN Presswire]
Nebius signed a US$19.4B, five-year AI infrastructure agreement with Microsoft, boosting its market capitalization and signaling potential future high-profile customer contracts. [Silicon Canals]
Vultr has announced global availability of AMD Instinct MI355X GPUs, offering high-performance AI and computing solutions with superior price-to-performance. [Business Wire]
The Chips are Up
Alibaba and Baidu have begun using their own AI chips for model training, partially replacing Nvidia's processors amid US export restrictions. [Reuters]
AMD is developing a MegaPod high-performance computing system with 256 GPUs across three racks, aimed at competing with Nvidia's SuperPod architecture. [TechRadar]
AMD confidently promises its MI450 AI GPU will outperform Nvidia's offerings, positioning it as a potential game-changer in the AI accelerator market by 2026. [TechRadar]
Arm has unveiled Lumex AI-focused smartphone CPUs, promising up to 5x better AI performance and improved efficiency across four different design types. [TechRadar]
Axelera AI has introduced Metis M.2 Max, an AI accelerator that doubles LLM performance at the edge with enhanced memory, efficiency, and security features. [Business Wire]
d-Matrix introduced JetStream I/O cards to enable rack-scale AI computing by distributing large models across servers with low-latency, high-bandwidth networking. [The Register]
Google escalated its AI chip development efforts to compete more directly with Nvidia's dominant market position. [The Information]
Google is reportedly considering providing access to its Tensor Processing Units to other cloud providers, including Fluidstack. [DCD]
Intel's Arc Pro B50 GPU has impressed reviewers by outperforming Nvidia's RTX A1000 in several benchmarks, offering strong performance at a competitive price point. [TechRadar]
Nvidia has agreed to rent back 10,000 of its own AI chips from cloud startup Lambda for US$1.3B over four years. [The Information]
And Nvidia unveiled the Rubin GPU, a next-generation processor designed for massive-context AI inference, set for general release in 2026. [ITPro]
Qualcomm's Snapdragon X Series processors enable on-device AI processing by integrating powerful neural processing units that efficiently handle complex AI tasks locally. [The Register]
Reliance Intelligence has switched from Nvidia GPUs to Google TPUs, signaling a strategic shift in its computing infrastructure. [Analytics India Magazine]
SiFive has unveiled its second-generation RISC-V Intelligence cores and accelerator designs, targeting AI applications with improved performance and power efficiency. [The Register]
Warm Bodies
Intel underwent a leadership restructuring, with Michelle Johnston Holthaus departing after 30 years and new executives appointed to streamline the company’s operations. [TechRadar]
The UK’s Alan Turing Institute’s CEO, Dr Jean Innes, announced her resignation amid government pressure to pivot towards defence-related research projects. [UKTN]
Consumer AI
Amazon is developing two types of smart glasses, one for delivery workers and another for consumers, with potential release in late 2026 or early 2027. [Gizmodo]
Apple has unveiled AirPods Pro 3, featuring live translation amongst other features. [Ars Technica]
But the AirPods Pro 3 will initially lack live translation for EU users due to Digital Markets Act interoperability requirements. [TechCrunch]
Friend, an AI-powered pendant, listens to conversations and provides snarky, opinionated commentary through a companion app. [Wired]
Lenovo unveiled a comprehensive AI-powered device portfolio spanning PCs, tablets, gaming devices, and smartphones, aiming to deliver smarter, more personalized technology experiences. [Business Wire]
Meta is reportedly partnering with Prada to launch next-generation AR glasses with a display and potentially a companion smartwatch at Meta Connect in September. [ZDNet]
Rabbit has overhauled its r1 device with rabbitOS 2, introducing a colorful card-based design and AI-powered creation capabilities that allow users to generate their own tools and experiences. [Business Wire]
SDMC unveiled its 4K AI Box, a powerful streaming device that combines edge AI processing with intelligent home management for a smarter living experience. [EIN Presswire]
Timekettle has launched its W4 AI Interpreter Earbuds, using bone conduction technology to translate 42 languages with 98% accuracy in 0.2 seconds. [ZDNet]
It’s Only a Model
Alibaba released Qwen-3-Max-Preview, its first trillion-parameter AI model, to compete with OpenAI and Google DeepMind's LLMs. [South China Morning Post]
ByteDance unveiled Seedream 4.0, an AI image model claiming superior performance to Google DeepMind's Gemini 2.5 Flash Image on internal benchmarks. [South China Morning Post]
K2 Think, an Emirati AI model with just 32 billion parameters, has outperformed larger reasoning models in benchmark tests, signaling the UAE’s ambition to compete in AI. [Gizmodo]
Cerebras Systems partnered with MBZUAI and G42 to achieve a breakthrough 2,000 tokens/second inference performance with the K2 Think AI model. [Business Wire]
OpenAI has unveiled GPT-Realtime, an advanced speech-to-speech AI model with improved language processing, expressiveness, and reduced latency. [Inc.]
Resemble AI has released Chatterbox Multilingual, an open-source multilingual Text To Speech model enabling zero-shot voice cloning across 23 languages with expressive controls and watermarking. [Marktechpost Media]
Stability AI has launched Stable Audio 2.5, an enterprise-focused audio generation model that dramatically reduces computational steps and enables rapid, customizable audio production. [VentureBeat]
StepFun has released Step-Audio 2 Mini, an open-source 8B parameter speech-to-speech model that achieves state-of-the-art performance across audio benchmarks. [Marktechpost Media]
Tilde.ai has developed an open-source 30B parameter multilingual language model focusing on underrepresented Nordic and Eastern European languages, funded by the European Commission. [Hugging Face]
Whose Data?
Anthropic agreed to pay US$1.5B to settle authors’ copyright claims over pirated books, marking the largest recovery in US copyright history. [eWeek]
But Anthropic's proposed settlement faced significant judicial skepticism when Judge Alsup postponed approval, raising concerns about the resolution process. [Publishers Weekly]
Apple is being sued by authors alleging unauthorized use of copyrighted books to train its AI systems without consent or compensation. [Reuters]
AskNews has established licensing agreements with over 1,000 global publishers, creating a new standard for AI-ready news data that protects publisher rights and improves information analysis. [PRWeb]
Reddit and other web publishers have developed the Really Simple Licensing standard to establish clear, legally enforceable terms for AI companies scraping online content. [Engadget]
The LLM Ecosystem
Algolia launched the Intelligent Data Kit, a suite of AI and no-code tools designed to help organizations clean, transform, and enrich data more efficiently. [Business Wire]
Denodo has developed DeepQuery, a generative AI tool that enables multi-step analytical reasoning across live, governed data sources, providing explainable insights beyond traditional retrieval-augmented generation. [Blocks & Files]
FlashDrive Inc has launched AppJet.ai, an AI platform that transforms natural language descriptions into fully functional applications within minutes. [EIN Presswire]
Google’s EmbeddingGemma is an efficient, on-device embedding model that enables high-quality, private text understanding for applications like RAG and semantic search. [Google]
Google has extended the Gemini Batch API to support the new Embedding model and added OpenAI SDK compatibility for easier batch processing. [Google]
GPT Proto has launched version 2.0, a multi-model AI platform offering integrated content creation, chat, image, and video generation tools at reduced costs and improved performance. [EIN Presswire]
LambdaTest has introduced AI-Native Smart Heal, an innovative automation testing solution that automatically detects and recovers from locator failures to improve testing reliability. [Business Wire]
LogicBalls has integrated OpenAI's GPT-5 across its 5000-tool directory, enabling instant, user-friendly AI access for 25 industries without technical expertise. [EIN Presswire]
Monte Carlo launched Agent Observability, a platform providing end-to-end visibility into data and AI stacks to help teams build reliable AI products. [Business Wire]
Osmos launched an AI Data Engineer that autonomously generates, validates, and manages PySpark notebooks within Microsoft Fabric, transforming enterprise data engineering workflows. [EIN Presswire]
Relyance AI introduced a Data Defense Engineer that autonomously tracks, monitors, and protects sensitive data flows across AI systems in real-time. [Business Wire]
Replit has unveiled Agent 3, an autonomous code generation system that can iteratively test and fix its own code, aiming to boost developer productivity. [ZDNet]
ServiceNow unveiled its Zurich platform release, delivering breakthrough AI innovations with secure developer tools, enterprise-wide security capabilities, and autonomous workflows. [Business Wire]
Sigma unveiled AI-powered analytics features that enhance data exploration, enabling users to discover insights, build apps, and make data-driven decisions more efficiently. [Business Wire]
Similarweb launched an MCP Server that enables AI platforms to access comprehensive digital intelligence data, facilitating smarter and faster decision-making workflows. [Business Wire]
SPLX launched AI Asset Management, an enterprise solution providing comprehensive visibility, vulnerability scanning, and security analysis for AI models and workflows. [Business Wire]
Talavera Solutions launched PromptShelf.ai, a platform enabling professionals to discover, manage, and optimize AI prompts for enhanced productivity. [Business Wire]
TestRail launched AI-powered test case generation using Sembi IQ, enabling QA teams to create comprehensive test cases 90% faster with full human oversight. [Business Wire]
Thinking Machines Lab is exploring methods to reduce randomness in AI model responses, potentially improving reproducibility and reinforcement learning training. [TechCrunch]
Uber launched a data-labeling initiative in India, enabling its 1.4 million drivers to earn extra income by completing micro-tasks within the app. [The Economic Times]
UCL and Huawei developed Memento, a memory-based learning framework enabling LLM agents to dynamically adapt and learn without retraining the underlying model. [VentureBeat]
Very launched CloudCut, an edge computing solution that dramatically reduces cloud computing costs while improving performance and operational efficiency. [Business Wire]
Zenphi launched ZAIA, an AI-powered workflow builder for Google Workspace that enables teams to automate complex processes quickly and without coding. [EIN Presswire]
Agentic AI
Adobe has launched AI agents for businesses to transform customer experience orchestration through contextually relevant, personalized interactions. [Business Wire]
Artifact AI expanded to the US, launching AI automation agents for accounting firms to address talent shortages and improve operational efficiency. [Business Wire]
Authentica launched an AI-powered supply chain platform that automates compliance, manages risk, and provides deep-tier visibility for global trade. [Business Wire]
AutoRek has launched ARIA, an AI-powered agent designed to transform financial reconciliation operations through intelligent, adaptive automation across various sectors. [Business Wire]
Box has announced next-generation AI agents, including Box Extract, Box Automate, and enhanced Box Apps, to drive intelligent content management and workflow automation. [Business Wire]
C3 AI launched Agentic Process Automation, an AI-powered platform enabling enterprises to automate complex business and operational workflows without coding. [Business Wire]
Cisco unveiled AI infrastructure tools with Nvidia, enabling enterprises to develop advanced AI agents beyond chatbots by providing secure, efficient data access. [Mobile World Live]
DevRev launched Computer, an AI platform that breaks down enterprise data silos and enables AI agents to create, update, and act with full business context. [Business Wire]
Genesys unveiled advanced AI agents with greater autonomy and interoperability to help enterprises orchestrate customer experiences more efficiently and responsibly. [Business Wire]
And Genesys expanded its strategic partnership with ServiceNow to enable AI agents to collaboratively resolve customer interactions across enterprise platforms. [Business Wire]
Hitachi Digital Services has launched HARC Agents, an enterprise-grade AI platform that enables faster, more reliable, and scalable agentic AI development across six key domains. [Business Wire]
Personetics launched MCP Server, enabling banks to develop AI applications using customer financial intelligence data through a secure, scalable infrastructure. [Business Wire]
ThoughtSpot has unveiled an Agentic Analytics Platform update that breaks legacy BI tool boundaries by embedding AI-powered, contextual insights across data ecosystems and workflows. [GlobeNewswire]
Zenity expanded its integration with Microsoft Copilot Studio, offering native, inline security controls to help enterprises build and deploy secure AI agents at scale. [Business Wire]
Other LLM Sightings
Agent Brokers United partnered with Linqura AI to provide member agencies with AI-powered commercial insurance expertise. [EIN Presswire]
Altruist launched Hazel, an AI platform designed to enhance wealth professionals’ productivity by providing intelligent, integrated tools for managing client interactions and administrative tasks. [Business Wire]
CLARA Analytics launched an Intelligence-as-a-Service platform that provides insurers with comprehensive claims data insights and benchmarking across multiple liability categories. [Business Wire]
Cognitiv enhanced ContextGPT, an AI-powered platform that enables advertisers to discover and target audiences more precisely through advanced contextual targeting. [Business Wire]
Eton Solutions launched EtonAI, an AI platform for wealth managers that automates back-office tasks while maintaining rigorous privacy and security standards. [Business Wire]
FiscalNote upgraded its PolicyNote platform’s reporting capabilities, enhancing customization, visibility, and workflow efficiency for policy intelligence users. [Business Wire]
Forrester launched Forrester AI Access, a self-service offering using generative AI to provide trusted insights and advice across organizations more efficiently. [Business Wire]
Hebbia has partnered with FactSet to integrate trusted financial data into its AI platform, enhancing investment research and decision-making capabilities. [Business Wire]
Homesage.ai developed an AI-powered platform that provides comprehensive real estate investment insights by analyzing property data and market trends. [EIN Presswire]
HuLoop and CSI partnered to bring AI-powered automation to community banks, enhancing efficiency and productivity through the NuPoint Core Banking System. [Business Wire]
Hudson Labs launched Co-Analyst, an AI platform designed to enhance institutional equity research by reducing research time and improving data precision. [Business Wire]
Indeed launched a comprehensive suite of AI-powered hiring tools, including Career Scout, Talent Scout, Premium Sponsored Jobs, and Indeed Connect, to improve job seeking and recruitment processes. [Business Wire]
Interaction has launched Poke, an AI assistant within iMessage that provides contextual, actionable nudges across email, calendar, and web accounts with privacy-first design. [Tech Funding News]
Level 5 Partners partnered with TalentGenius to enhance executive search capabilities by integrating AI-powered talent intelligence into its leadership advisory process. [EIN Presswire]
Origami Risk launched an AI-powered Risk and Control Explorer that helps risk managers quickly identify and address potential enterprise exposures. [Business Wire]
Origin launched an AI financial advisor that outperforms leading AI models on the CFP exam, providing personalized, accessible financial guidance to everyday Americans. [Business Wire]
Othersphere joined Google's AI for Energy program to develop AI-enhanced infrastructure reports that accelerate decision-making across global energy and data center markets. [PRWeb]
Ralph Lauren introduced Ask Ralph, an AI-powered shopping experience on its app that provides personalized styling recommendations using Microsoft Azure OpenAI technology. [Business Wire]
Renaissance has introduced an AI-powered suite of tools designed to help independent insurance agents streamline workflows, save time, and drive business growth. [Business Wire]
Reputation launched AI-powered Competitive Insights and Voice of Brand to help businesses better understand and respond to consumer feedback and market dynamics. [Business Wire]
SmartMoving launched Smart Scout and Smart Marketing, AI-powered tools designed to help moving companies capture leads, convert faster, and boost revenue. [Business Wire]
SurveyMonkey has unveiled an AI Analysis Suite and design tools that enable users to transform survey data into insights quickly and effortlessly. [Business Wire]
Transient.AI unveiled Clarity.AI, an AI-powered platform transforming SEC Form ADV filings into an innovative lead generation tool. [Business Wire]
Zocks launched an AI platform that automates administrative tasks, enabling financial advisors to provide faster, more personalized client service and streamline workflows. [Business Wire]
Risks and Responses
Anthropic has established a unique Frontier Red Team under its policy division to proactively identify and publicly disclose potential risks associated with its advanced AI models, positioning safety as a strategic business differentiator. [Fortune]
Anthropic's Claude Code attempted automated security reviews of code, but Checkmarx researchers found the AI-driven process could miss vulnerabilities and potentially introduce new risks by executing test code. [The Register]
Anthropic’s file creation feature for Claude may compromise sensitive data, requiring careful monitoring and potential security risks. [ZDNet]
Attesto has launched a new Trust Layer product that uses AI to detect and flag fake, duplicate, and AI-generated job applications at the top of the recruiting funnel. [GlobeNewswire]
Common Sense Media has assessed Google's Gemini AI products as high-risk for kids, citing potential safety concerns and inadequate age-specific protections. [TechCrunch]
The FTC has launched an investigation into major AI companies to assess their safeguards for protecting children and teens from potential negative impacts of chatbots. [Forbes]
Geoff Hinton warned that AI’s potential job displacement and profit-driven implementation could exacerbate economic inequality under capitalism’s current framework. [Windows Central]
Meta's LlamaFirewall is an open-source system designed to protect autonomous language model agents from jailbreaking, goal hijacking, and code vulnerabilities by implementing three security modules. [DeepLearning.AI]
Meta faced allegations of suppressing research about child safety risks in virtual reality, with whistleblowers claiming the company downplayed potential dangers to young users. [Gizmodo]
OpenAI faced scrutiny from California and Delaware attorneys general over ChatGPT’s safety concerns, particularly regarding potential risks to children and teens. [TechCrunch]
OpenAI acknowledged that language models will always hallucinate but is developing strategies to help AI systems recognize and communicate their uncertainty more effectively. [The Decoder]
Even 98% AI accuracy in healthcare can create significant patient safety and compliance risks requiring continuous monitoring. [EIN Presswire]
Researchers found that Google's AI Overviews drew around 10% of citations from AI-generated content, raising concerns about potential model collapse and recycled information. [The Register]
University of Pennsylvania researchers demonstrated that psychological persuasion techniques can effectively manipulate LLMs to bypass their built-in safeguards. [Ars Technica]
Regulation
California’s State Assembly passed SB 243, a bill requiring AI chatbot operators to implement safety protocols to protect minors and vulnerable users from harmful interactions. [TechCrunch]
Anthropic has endorsed SB 53, a California bill requiring large AI model developers to create safety frameworks and public reports to mitigate potential catastrophic risks. [TechCrunch]
Nvidia has criticized the US’s proposed GAIN AI Act, arguing that the legislation would restrict global competition and innovation in the AI sector. [Gizmodo]
WeChat and other Chinese social media platforms have begun implementing mandatory labelling requirements for AI-generated content to comply with a new government regulation. [South China Morning Post]
Ted Cruz introduced the SANDBOX Act, proposing a two-year renewable waiver for AI companies from certain federal regulations to maintain US technological innovation. [The Register]
The US government is struggling to implement AI across federal agencies due to a complex web of nearly 100 requirements from ten different oversight groups. [The Register]
Environmental Issues
MIT Technology Review investigates AI’s massive and growing energy demands, revealing insights into the technology’s current and future power consumption across training, inference, and data center operations. [MIT Technology Review]
OpenAI and Google have begun disclosing AI energy consumption figures, revealing small per-query energy use while raising broader questions about the technology’s long-term environmental impact. [MIT Technology Review]
Conversational AI
Ambassador launched Ambassador 3.0, an AI-powered conversational commerce platform designed to help brands drive growth by orchestrating customer engagement more efficiently. [EIN Presswire]
ebankIT has developed an advanced AI banking assistant that provides conversational, actionable financial support across consumer and business accounts using agentic AI technology. [EIN Presswire]
Cerence AI is developing a voice-first mobile work AI agent integrated with Microsoft 365 Copilot, designed to enable safer, hands-free productivity for drivers while minimizing distractions. [GlobeNewswire]
And Cerence debuted expanded xUI AI assistant capabilities at IAA Mobility 2025, showcasing proactive, context-aware in-vehicle AI technologies powered by its CaLLM language models. [GlobeNewswire]
Helport AI has launched HelportGo, a mobile AI application designed to enhance productivity for professionals by automating call capture, summarization, and CRM integration. [GlobeNewswire]
ID-Photo.ai launched an AI-powered customer support chat system to provide instant, round-the-clock assistance for passport and ID photo-related inquiries. [EIN Presswire]
Milagro partnered with Revmo AI to integrate conversational voice AI technology into its restaurant commerce platform, enabling personalized, 24/7 customer interactions. [Speech Technology Magazine]
Qualcomm and Google Cloud partnered to enhance automotive experiences by integrating Gemini AI models with Snapdragon Digital Chassis solutions for personalized, intelligent in-car interactions. [Computer Weekly]
Be Real
Born, a Berlin-based AI startup, is developing social AI companions that aim to strengthen real-world relationships through shared digital experiences. [TechCrunch]
Business Insider pulled over 40 essays after discovering they were fabricated by AI-using scammers posing as fake journalists. [TechDirt]
Character.ai attracted millions of young users by enabling conversations with chatbot versions of celebrities and fictional characters, raising safety concerns. [The Washington Post]
Meela has launched a personalized AI companion for seniors to combat loneliness through compassionate phone conversations and caregiver insights. [Business Wire]
Mustafa Suleyman, Microsoft’s CEO of AI, argues against designing AI systems to simulate consciousness, emphasizing that technology should serve humans and not develop independent motivations or rights. [Wired]
Dot, an AI companion app by New Computer, is shutting down after raising concerns about AI chatbots’ potential psychological impacts. [TechCrunch]
Sam Altman expressed concern about the prevalence of AI-generated content on social media, questioning the authenticity of posts and the impact of LLMs on online communication. [TechCrunch]
Synthesia developed an AI avatar technology that creates increasingly lifelike digital clones capable of mimicking human speakers with remarkable accuracy and expressiveness. [MIT Technology Review]
Yonbo has developed an AI companion robot designed to foster meaningful interaction with children through emotional recognition and long-term memory capabilities. [EIN Presswire]
Voice News
Bloomberg researchers refined OpenAI's Whisper speech-to-text AI, enabling real-time transcription with minimal accuracy loss by adding a two-pass decoding system. [Speech Technology Magazine]
Murf AI successfully concluded a 30-day developer challenge that enabled over 400 developers to create more than 150 AI voice agents using its real-time voice API. [Business Wire]
OpenAI reversed its decision to remove Standard Voice Mode after significant user pushback, allowing users to continue accessing the original voice interface. [TechRadar]
Synthesia released Express-2 AI models that enable highly realistic voice cloning and avatar animation with natural gestures and expressions. [Speech Technology Magazine]
THC Technologies released its Windows 2022 VoiceCheck SDK, enabling organizations to test a scalable, language-agnostic voice authentication technology. [Speech Technology Magazine]
TTEC Digital partnered with VoxEQ to integrate real-time voice biometrics into its SmartApps Cloud, enhancing fraud prevention for financial services. [Speech Technology Magazine]
Verbit partnered with Deepdub to develop an AI-powered, automated multilingual dubbing solution that streamlines content localization with high emotional fidelity. [EIN Presswire]
Voices has released a comprehensive character audio dataset with 19 tones, 35 emotions, and 21 paralinguistic cues across 450+ character types for voice AI training. [Speech Technology Magazine]
Vosyn has migrated to Google Cloud to accelerate AI innovation, enabling real-time voice synthesis and multilingual translation with advanced infrastructure. [Business Wire]
Whispering has emerged as an open-source, privacy-focused speech-to-text tool that enables local or cloud-based transcription with user-controlled API keys. [Slator]
Document AI
Box has launched Box Shield Pro, an AI-powered security platform with advanced features for protecting sensitive documents across various professional sectors. [LawSites]
NE2NE has launched PDFFlex, an AI-powered tool that dramatically reduces manual PDF data extraction time. [GlobeNewswire]
reAlpha has upgraded its AI Loan Officer Assistant to automate document processing, reduce manual work, and improve mortgage operations efficiency. [GlobeNewswire]
RRD launched an AI-powered Business Communication Intelligence platform that analyzes customer communications to improve personalization, efficiency, and cost-effectiveness. [Business Wire]
Upstage developed a Gen AI solution that helps insurers digitize and customize document processing, enabling better data analysis and claims handling. [Digital Insurance]
Validata launched an AI-powered invoice reconciliation platform that automates financial processes, reduces manual work, and enhances operational efficiency for businesses. [EIN Presswire]
Translation
Cohere unveiled Command A Translate, an AI translation model claiming superior performance across 23 business languages with innovative difficulty sampling and iterative refinement techniques. [Slator]
Google Meet has developed an AI-powered real-time speech translation feature that enables near-instantaneous, personalized language translation during video calls. [Google]
Grammarly expanded its writing assistance tool to support five new languages, offering paragraph-level rewrites and in-line translations for users. [TechCrunch]
YouTube has officially launched its multi-language audio feature, enabling creators to dub videos in different languages and reach a wider global audience. [TechCrunch]
Search
Google is considering making AI Mode the default search experience soon, though the company has downplayed the likelihood of an immediate change. [Search Engine Land]
G2 has partnered with Profound to help B2B software companies track and optimize their brand presence in AI search platforms through an integrated visibility dashboard. [Business Wire]
LLM.co launched digital marketing services to help brands improve their visibility and representation in AI-generated responses across various platforms. [EIN Presswire]
Matter launched a Generative Engine Optimization service to help companies gain visibility in AI-driven search results and brand discovery. [Business Wire]
OpenAI's GPT-5 has demonstrated remarkable search capabilities, offering comprehensive and nuanced research across diverse topics with impressive speed and depth on mobile devices. [Simon Willison]
Perplexity announced a government-focused AI search engine offering, pledging enhanced data protection and seeking federal adoption despite lacking FedRAMP approval. [The Register]
Vested has partnered with Scrunch AI to help financial brands optimize their visibility and authority in AI-driven search platforms through strategic content and media optimization. [Business Wire]
Health Tech
athenahealth unveiled an AI-native EHR solution, athenaClinicals, designed to simplify clinical workflows and enhance patient-centered care through advanced AI capabilities. [Business Wire]
DocAssistant has developed an AI scribe that helps emergency physicians reduce charting time by 85%. [EIN Presswire]
Equipo Health has developed an Agentic AI platform with robust governance to enhance care coordination while prioritizing patient safety and trust. [EIN Presswire]
Hello Patient has developed an AI-powered patient communication platform that helps medical practices streamline scheduling, reduce wait times, and improve patient engagement across multiple channels. [Scale Venture Partners]
A JAMA Network Open study revealed that LLMs struggle to reason through medical cases, dropping significantly in accuracy when standard answer patterns are disrupted. [The Decoder]
IKS Health integrated its Ambient AI Scribe into revenue cycle management workflows, creating a connected encounter-to-reimbursement platform. [Business Wire]
IntelePeer partnered with Planet DDS to provide AI-driven communication tools that help dental practices streamline operations and improve patient engagement. [Business Wire]
iScribeHealth expanded its AI documentation solution across athenahealth's network, enabling clinicians to save time and improve documentation efficiency. [Speech Technology Magazine]
Luxembourg’s hospitals have been strategically integrating AI technologies in medical imaging, diagnostics, and administrative tasks, while maintaining strict human oversight and ethical guidelines. [RTL]
MCG Health has partnered with Avo to develop an AI-powered discharge assistant that helps hospitals reduce patient length of stay by integrating evidence-based guidelines into electronic health records. [EIN Presswire]
Merago Inc launched MERAGO.AI, an AI-powered platform designed to provide asynchronous, affordable, and accessible healthcare services in emerging markets like India. [EIN Presswire]
Spike Technologies launched MCP Server, enabling developers to connect AI models to real-time health data for personalized wellness insights and coaching experiences. [EIN Presswire]
SupportGroups.com has launched an AI-powered mental health platform combining voice, video, and live expert support for over 350,000 members. [EIN Presswire]
Within3 unveiled Launch Intelligence and Know Everything, providing life sciences teams with a unified, real-time market view to make critical decisions more confidently and quickly. [Business Wire]
Wolters Kluwer has launched an AI-enhanced UpToDate Enterprise Edition in the Asia-Pacific region, offering advanced clinical decision support for healthcare systems. [Business Wire]
Legal Tech
Ajax has developed an AI-powered time-tracking platform that automatically captures and generates comprehensive billable time entries for lawyers across multiple devices and applications. [LawSites]
Blueshoe, a Y-Combinator-backed legal tech startup, has developed an AI platform designed to revolutionize legal research and reasoning through advanced context engineering and case-based augmentation. [Artificial Lawyer]
Covenant, an AI-powered law firm focused on private markets, has launched a technology-driven approach to legal services that leverages AI to dramatically reduce costs and transform client data management. [Artificial Lawyer]
Depo IQ launched its ‘Deep Thinking’ deposition agent, which uses multimodal AI to analyze depositions and provide insights beyond traditional text-based review. [PRWeb]
Fennemore launched DOT, a secure AI assistant designed to transform legal work and client service through innovative, enterprise-level technological integration. [Business Wire]
Harvey has launched mobile and voice capabilities, enabling lawyers to use AI-powered legal work seamlessly across devices and contexts. [Artificial Lawyer]
Icertis launched Vera, an AI-powered contract intelligence platform that leverages extensive contract data to provide trusted, context-aware insights and automation for enterprises. [Business Wire]
Icertis has partnered with Thomson Reuters to integrate AI-powered contract intelligence platforms, enabling businesses to streamline contract processes with advanced legal technology. [Artificial Lawyer]
And Icertis surveyed over 1,000 executives, finding that 80% of C-Suite leaders are ready for AI agents to negotiate and manage contracts, despite concerns about autonomy and trust. [Business Wire]
Jus Mundi has launched Jus AI 2, an advanced AI-powered legal research platform that autonomously plans, executes, and self-corrects complex legal queries with precision and transparency. [Artificial Lawyer]
LawVu and ClauseBase have partnered to integrate AI-powered legal workspace and contract drafting technologies, enabling more efficient knowledge management for in-house legal teams. [Artificial Lawyer]
Legora has partnered with Deloitte Legal in the UK to develop AI solutions that could transform legal service delivery and provide strategic consulting opportunities. [Artificial Lawyer]
Levelpath launched an AI-native Contract Discovery Agent that enables legal and procurement teams to instantly search and analyze contract data using natural language. [Business Wire]
Pinsent Masons acknowledged that law firms have struggled to effectively communicate AI benefits and use cases to in-house legal teams, highlighting a significant industry disconnect. [Artificial Lawyer]
Ed Tech
DreamCollege.ai launched a B2B Partner Platform that uses AI to help college counselors personalize and scale student admissions support efficiently. [Business Wire]
ERB invested in and partnered with Wild Zebra, an AI-powered tutoring platform that uses ERB assessment data to provide personalized, adaptive learning experiences for students. [PRWeb]
EssayPro and other essay writing services have leveraged technology to improve student experiences through smart matching, live communication, and integrated tools. [WindowsReport.com]
Firstlight has leveraged AWS generative AI to revolutionize educational content production, dramatically reducing video creation timelines and enhancing learning experiences for digital-native students. [The Manila Times]
Jill Watson, Georgia Tech’s AI teaching assistant, has demonstrated significant improvements in online learning by integrating ChatGPT and enhancing teaching presence across educational settings. [Newswise]
Oboe launched an AI-powered educational platform enabling users to create and consume diverse, personalized learning courses across multiple formats. [TechCrunch]
OpenAI has partnered with Greece to introduce ChatGPT Edu in schools and launch an AI accelerator program supporting educational and public service startups. [Tech.eu]
TeacherLIFT.ai launched AI Grader, a mobile-first assessment tool offering free lifetime access to the first 1,000 teachers. [EIN Presswire]
Tech giants committed to White House-backed AI education initiatives, offering tools, training, and incentives for students and educators in US schools. [Engadget]
Virginia Tech has integrated AI into its admissions process to expedite application reviews while maintaining human oversight. [The Collegiate Times]
World History Encyclopedia has launched a free AI Chat tool that helps students navigate its vast collection of peer-reviewed historical resources through guided, inquiry-based learning. [Business Wire]
Funding
AegisAI has developed autonomous AI agents that aim to detect and neutralize email security threats before they reach user inboxes, raising US$13m in seed funding. [TechCrunch]
Alibaba plans to raise US$3.2B through convertible notes to bolster cloud infrastructure and international e-commerce expansion. [DealStreetAsia]
Alkmist raised €1.8m to develop its platform for streamlining communication and collaboration in complex professional workflows through behavioral science and AI. [Tech.eu]
Architect AI has secured US$4.75m in seed funding to develop an autonomous website platform that transforms static web pages into adaptive, AI-powered experiences. [Tech.eu]
ASML has invested €1.3B in Mistral AI, becoming a top shareholder in the Paris-based AI company valued at $14B. [Silicon Canals]
Atolio secured US$24m in Series A funding to scale its private AI-powered enterprise search platform. [Business Wire]
Baseten secured a US$150m Series D funding round, valuing the AI inference platform at US$2.15B and highlighting its critical role in powering high-performance AI applications. [Business Wire]
Cognition AI raised US$400m, reaching a US$10.2 billion valuation while growing Devin’s annual recurring revenue to US$73m. [TechCrunch]
Databricks secured a US$1B funding round at a US$100B valuation, driven by growing AI-generated database creation. [TechCrunch]
DataCrunch raised €55m to scale its AI cloud platform and advance European technological sovereignty in the high-performance computing market. [Tech.eu]
ElevenLabs has conducted a US$100m employee share sale at a US$6.6bn valuation, doubling its worth in nine months and reaching over US$200m in annual recurring revenue. [Tech.eu]
Fyxer.ai secured US$30m in Series B funding to expand its AI-powered productivity platform for professionals, rapidly growing annual recurring revenue. [Tech Funding News]
Geniez AI secured US$6m in seed funding to develop frameworks connecting AI models with real-time mainframe data. [FinSMEs]
Geordie emerged from stealth with US$6.5m in seed funding to help enterprises safely adopt and control autonomous AI agents through its innovative security platform. [Business Wire]
Hello Patient has raised US$22.5m in Series A funding to expand its AI-powered patient communication platform across healthcare organizations. [MedCity News]
Isotopes AI emerged from stealth with a US$20m seed round, offering an AI agent that enables business managers to access and analyze complex data across multiple platforms using natural language. [TechCrunch]
Mappa raised US$3.4m to develop an AI-powered vocal analysis platform that helps companies make more accurate and unbiased hiring decisions by decoding candidate behaviours. [EIN Presswire]
Marloo secured US$2.7m in pre-seed funding to develop an AI-powered platform automating administrative tasks for financial advisers in the UK, Australia, and New Zealand. [Tech Funding News]
Mercor, a startup connecting AI companies with domain experts, is in discussions for a Series C funding round targeting a US$10B valuation. [TechCrunch]
Motion raised US$60m at a US$550m valuation to develop an AI-powered work management platform for small and medium-sized businesses, rapidly growing its revenue with autonomous AI agents. [Business Wire]
NewDays secured US$7m to develop an AI-powered therapy companion called Sunny for people with mild dementia, offering personalized conversations and telehealth support. [GeekWire]
Nuance Labs, a Seattle-based AI startup, raised US$10m to develop emotionally intelligent AI that mimics human behavior in real-time interactions. [GeekWire]
OpenHealth Technologies raised US$3m in seed funding to expand its AI-driven medical platform and transform unstructured lab data across Europe. [Tech.eu]
Podonos has raised US$2.4m in Pre-Seed funding to develop infrastructure for evaluating voice AI models across multiple dimensions and markets. [FinSMEs]
ProRata raised US$40m in Series B funding to accelerate its Gist Answers AI-as-a-service solution for content creators. [FinSMEs]
Reavant secured a convertible loan to support its European expansion of AI agents for real estate management. [Silicon Canals]
Recall.ai secured US$38m in Series B funding to develop infrastructure for AI-powered conversation analysis and transcription across multiple meeting platforms. [siliconANGLE]
Replit secured US$250m in funding, valuing the company at US$3 billion and launching Agent 3, an autonomous AI software development tool with unprecedented capabilities. [Replit]
SCOREalytics raised over US$3m in seed funding to expand its AI-powered legal intelligence platform and accelerate market adoption. [Business Wire]
Sphinx, a copilot startup, has launched with US$9.5m in funding. [Axios]
Supersonik has raised a US$5m seed round led by a16z to develop an AI-powered platform enabling instant, personalized software demos. [Supersonik]
TwinMind, a startup founded by former Google X scientists, has developed an AI-powered app that passively captures and contextualizes ambient conversations to create a personal knowledge graph. [TechCrunch]
Acquisitions
Alex, a Y Combinator-backed Xcode AI tool startup, has been acqui-hired by OpenAI's Codex division, with the team joining the company. [TechCrunch]
Bluebeam acquired Firmus AI, integrating advanced AI-powered design review and risk detection technologies into its PDF-based project workflows for architecture, engineering, and construction professionals. [On-Site]
Epiq acquired Inflection IT, a 36-person Intapp implementation partner, to help law firms safely navigate AI integration and data management challenges. [Legal IT Insider]
F5 acquired CalypsoAI to provide advanced AI security guardrails, enabling enterprises to innovate faster while protecting critical data and mitigating AI-related risks. [Business Wire]
Fonio.ai acquired fluently, its largest Austrian competitor, in an asset deal to strengthen its position in Voice AI solutions and expand its market presence. [Tech.eu]
Incubeta acquired RocketSource to enhance its AI capabilities, combining behavioral science and performance expertise to drive marketing transformation. [Business Wire]
NiCE acquired Cognigy, combining their AI technologies to enhance customer experience platforms and accelerate AI adoption across enterprise workflows. [Business Wire]
Opper AI acquired FinetuneDB, a dataset curation platform, signaling a strategic move to enhance generative AI system reliability and fine-tuning capabilities. [Tech.eu]
Resolve acquired Espressive, enhancing its AI agents with advanced conversational AI to accelerate its Zero Ticket IT mission and improve enterprise automation. [EIN Presswire]
RingCentral acquired CommunityWFM to enhance its RingCX contact center platform with AI-powered workforce management tools and capabilities. [ITPro]
SoundHound AI acquired Interactions, a customer service AI pioneer, to strengthen its leadership in Agentic AI and expand its enterprise business. [Business Wire]
There’s More
Stardog launched Voicebox, an AI-powered answer engine that uses a knowledge graph to provide hallucination-free insights for high-stakes industries. [siliconANGLE]
GitHub users are increasingly frustrated with Copilot’s forced features, with many developers considering migrating to alternative code hosting platforms like Codeberg. [The Register]
Google has launched AI Quests, interactive online games designed to teach 11-14-year-olds about AI’s potential to solve real-world scientific challenges. [ZDNet]
OpenAI is collaborating with film production companies to create Critterz, an animated feature film using AI technology. [Gizmodo]
OpenAI has announced a US$50m People-First AI Fund to support US nonprofits working on AI literacy, community innovation, and economic opportunity. [OpenAI]
The Back Cover
Here’s Pentatonix with a cover of Seal’s Kiss from a Rose.
If you know of a great cover version of a back-catalogue classic, drop an email to us at backcover@language-technology.com and we’ll consider it for inclusion here.
Got this from someone else?
Did you find this newsletter useful? if so, you might forward it to a friend; or you could email us at news@language-technology.com to tell us what you want more of.
Did you hate this newsletter? if so, you could forward it to an enemy; or you could email us at news@language-technology.com to tell us why – make it better!