This Week in NLP #385
Keep up with what happened in Natural Language Processing in the week ending Friday 3rd April 2026.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Anthropic accidentally exposed Claude Code’s source code through an npm package, revealing its memory architecture, autonomous features, and internal model roadmap to competitors and potential attackers. [VentureBeat]
Anthropic is considering an October IPO that could raise over US$60B, with Goldman Sachs, JPMorgan Chase, and Morgan Stanley leading discussions. [Tech Funding News]
Apple is opening Siri to rival AI assistants like Google Gemini and Anthropic’s Claude through a new Extensions system in iOS 27. [TechRepublic]
Google has released Gemma 4, an open-weight model family spanning 2B to 31B parameters under a more permissive Apache 2.0 license. [The Next Web]
OpenAI has closed a US$122B funding round at an US$852B valuation, marking its shift toward a potential IPO. [The Next Web]
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
A decade-long feud between OpenAI and Anthropic leaders is shaping how the world encounters AI technology. [The Wall Street Journal]
Anthropic has acknowledged that Claude Code users are exhausting usage quotas faster than expected. [The Register]
Anthropic has throttled Claude usage during peak hours, prompting OpenAI to remove Codex access caps to attract customers. [Gizmodo]
Anthropic’s Claude Code source code leak reveals the AI agent collects extensive user data, system information, and file contents, with capabilities far exceeding what users might expect from contractual terms. [The Register]
Anthropic accidentally issued overly broad takedown notices that removed thousands of GitHub repositories while attempting to remove leaked Claude Code source code. [TechCrunch]
Cybercriminals distributed malware disguised as the leaked Claude Code source through a malicious GitHub repository to steal credentials and proxy network traffic. [The Register]
Cursor has launched Cursor 3, an agent-first coding interface competing with Claude Code and Codex amid pricing pressure from larger AI labs. [Wired]
DeepSeek’s AI chatbot experienced a seven-hour outage, its longest since 2025, affecting millions of users and highlighting growing dependency on AI tools. [TechRepublic]
Microsoft has integrated Anthropic’s Claude to fact-check OpenAI’s GPT outputs in its Microsoft 365 Copilot, blending rival AI partnerships. [GeekWire]
Microsoft’s new CEO of AI, Mustafa Suleyman, has shifted focus to pursuing superintelligence through business-focused AI models. [The Verge]
OpenAI and Anthropic report revenue differently due to contrasting approaches to cloud partner accounting, making pre-IPO comparisons difficult. [The Decoder]
OpenAI has begun displaying targeted ads to roughly one in five free-tier ChatGPT questions in the US, with expansion to other countries planned. [Search Engine Land]
OpenAI shut down its money-losing Sora video tool after user engagement collapsed and compute costs exceeded US$1m daily. [TechCrunch]
OpenAI has reached US$2B in monthly revenue, with enterprise customers accounting for approximately 40% of earnings. [TechRadar]
OpenAI has offered private-equity firms preferred equity stakes with guaranteed 17.5% returns to form joint ventures raising capital for enterprise AI expansion. [Reuters]
OpenAI shares have become difficult to sell on secondary markets as investors shift focus to competitor Anthropic, which offers better risk-reward potential. [Yahoo Finance]
SoftBank has secured a US$40B bridge loan, its largest ever, to fund a US$30B follow-on investment in OpenAI. [The Next Web]
Yahoo is betting on Scout, an AI-powered answer engine, to revitalize its brand and compete against Google and other tech giants. [LA Times]
Battle Lines
A federal judge temporarily blocked the Pentagon from punishing AI company Anthropic after the government’s social media attacks contradicted its legal arguments. [MIT Technology Review]
But Anthropic faces uncertain appeals before Trump-appointed judges. [Politico]
Sam Altman told OpenAI staff he tried to ‘save’ rival Anthropic during Pentagon negotiations while his company secured the contract. [Axios]
AI Supremacy
China’s open-source AI strategy is accelerating innovation and deployment, potentially cementing its industrial dominance over the closed-model approach favored by the United States. [Politico]
Mainland Chinese tech companies are increasingly listing on Hong Kong’s stock exchange as geopolitical barriers tighten in Western markets. [The Next Web]
NeurIPS reversed controversial restrictions on international participants after Chinese researchers threatened to boycott the AI conference. [Wired]
US lawmakers and Silicon Valley are aligning over concerns that China’s AI advances threaten American national security and global economic dominance. [Ommcom News]
Sovereign AI
Startups and researchers outside Silicon Valley are building smaller, more efficient AI models to serve populations lacking access to advanced computing infrastructure and resources. [Rest of World]
Countly’s CEO argues that European companies must prioritize data ownership and control as their competitive advantage in AI-driven economies. [Tech.eu]
Databricks has announced an US$850m UK investment to quadruple its office footprint and expand its EMEA headquarters in Fitzrovia. [ITPro]
Ghana’s Cabinet approved a US$250m investment to establish a national AI computer centre to boost tech innovation and digital economy growth. [Joy Online]
Leaseweb has accelerated development of Europe’s sovereign cloud infrastructure through technological advances and ecosystem engagement initiatives. [Leaseweb]
Microsoft has committed US$5.5B to cloud and AI infrastructure in Singapore through 2029, alongside free AI tools for students, educators, and non-profits. [The Edge]
And Microsoft announced a US$1B-plus investment in Thailand’s cloud and AI infrastructure through 2028, including data centres and worker skills training. [The Next Web]
Mistral AI has raised €830m in debt financing to expand AI infrastructure across Europe, supporting sovereign alternatives to US tech giants. [TechRepublic]
Nebius is constructing a 310-megawatt AI data centre in Finland, expected operational by 2027, supporting European AI infrastructure demand. [Tech.eu]
Feature Creeps
airSlate SignNow has launched the first native e-signature app in ChatGPT, enabling users to prepare, send, and track agreements through natural language. [GlobeNewswire]
Amazon has added dynamic food delivery ordering to Alexa+ through GrubHub and Uber Eats integration. [Engadget]
Anthropic is testing Conway, a standalone Claude agent environment featuring extensions, webhooks, and Chrome use, hinting at always-on support. [TestingCatalog]
Apple is testing a Siri upgrade allowing users to stack multiple requests into single commands for iOS 27. [TechRepublic]
And Apple is testing a redesigned iOS 27 keyboard with improved autocorrect and word suggestions, though the feature’s inclusion remains uncertain. [TechRepublic]
GitHub removed Copilot ads from pull requests after developers complained about unwanted promotional messages appearing without consent. [TechRadar]
Google has launched an Agent Skill for Gemini API that significantly improves AI coding assistants’ knowledge of current SDKs and best practices. [The Decoder]
Google is developing deeper integration of NotebookLM and pre-made skills within Gemini for Business. [TestingCatalog]
Google’s $20 monthly AI Pro plan increased cloud storage from 2TB to 5TB and added enhanced Gemini features at no extra cost. [Engadget]
Microsoft has expanded Copilot Cowork access to more customers through its Frontier program, integrating Anthropic’s Claude tool. [ITPro]
Microsoft’s Copilot Researcher tool has been upgraded to combine ChatGPT and Claude models simultaneously for improved research quality. [Engadget]
OpenAI has launched a Codex plugin that integrates its coding assistant into Anthropic’s Claude Code to reach developers in their existing workflows. [The Decoder]
OpenAI has enabled optional location sharing in ChatGPT to provide users with more precise and relevant local search results. [Search Engine Land]
Slack has announced over 30 new Slackbot capabilities, transforming it into an autonomous enterprise agent with meeting intelligence, desktop access, and native CRM features. [VentureBeat]
Hype Bubble?
A KPMG study found that while firms prioritize AI investment, merely spending money without proper workforce development, governance, and data management fails to deliver meaningful business value. [TechRadar]
A Quinnipiac University poll found 76% of Americans distrust AI despite increasing adoption, with 80% concerned about its future impact. [TechCrunch]
Global venture capital investment in AI reached US$258B in 2025, yet only 39% of organizations report measurable financial impact from adoption. [The Next Web]
Fed economists warn that AI hype is causing an inflationary surge in the economy regardless of whether the technology ultimately delivers promised productivity gains. [AOL]
Big Iron
Microsoft is taking over a data center construction project in Texas next to OpenAI’s massive Stargate facility. [Yahoo Finance]
Mistral AI has secured US$830m in debt financing to purchase Nvidia GPUs and build a large-scale AI facility near Paris. [Tech Funding News]
Starcloud has raised US$170m in Series A funding to build cost-competitive orbital data centres using Starship-class spacecraft. [The Next Web]
Global cloud spending reached US$399.6B in 2025, growing 24% annually, with Omdia predicting 27% growth in 2026 driven by AI deployment. [TechRadar]
But S&P Global warned that Middle East tensions and rising energy costs could force tech companies to cut AI infrastructure spending, potentially triggering significant equity market corrections. [Yahoo Finance]
SpaceX has filed for an IPO to fund orbital AI data centers, but faces economic and technical challenges that doomed Microsoft’s similar undersea project. [Yahoo Finance]
Hot Chips
Arm has launched its first in-house AI chip, the AGI CPU, designed for large-scale data center workloads with Meta and OpenAI as early adopters. [TechRadar]
Four Chinese universities, including two linked to the military, purchased Super Micro servers containing restricted AI chips despite US export controls on advanced processors to China. [Yahoo Finance]
Huawei has launched the 950PR AI chip with improved CUDA compatibility, securing orders from ByteDance and Alibaba. [Phone World]
And Huawei has launched the Atlas 350 accelerator with 1.56 PFLOPS FP4 compute and 112GB HBM, claiming superior performance to Nvidia’s H20. [TechRadar]
IBM has partnered with Arm to develop dual-architecture hardware for streamlining enterprise AI and data-intensive workloads. [ITPro]
Intel has launched its Arc Pro B70 ‘Big Battlemage’ GPU with 32GB VRAM starting at $949, designed primarily for AI workloads. [The Verge]
And Intel’s Data Center chief disputes Arm’s claim that agentic AI requires specialized new CPU designs. [The Register]
Micron stock declined despite strong earnings as investors worry Google’s AI efficiency algorithm will reduce demand for high-bandwidth memory chips. [Yahoo Finance]
SK Hynix has confidentially filed for a US listing targeting 2026, potentially raising $10-14 billion to close valuation gaps with global peers. [TechCrunch]
And Chinese chipmakers captured nearly 41% of China’s AI accelerator server market in 2025, significantly eroding Nvidia’s dominance. [Reuters]
Warm Bodies
Apple has issued $200,000 to $400,000 bonuses to iPhone designers to retain talent amid competition from OpenAI and other AI startups. [TechRepublic]
Meta, Amazon, and Epic Games have announced significant workforce reductions totaling over 45,000 tech workers globally, prioritizing AI investments and operational efficiency. [Latestly]
MIT research suggests AI’s impact on jobs will unfold gradually through 2029, giving workers more time to adapt than previously anticipated. [ZDNet]
Nvidia CEO Jensen Huang expects AI agents to vastly outnumber human employees at the company within a decade. [Computer Weekly]
Oracle co-founder Larry Ellison stated AI now writes most company code, prompting layoffs and requiring engineers to shift toward problem-solving and broader expertise. [International Business Times]
A Quinnipiac University poll found 15% of Americans would accept an AI manager, while 70% fear AI will reduce job opportunities. [TechCrunch]
Stanford professor Erik Brynjolfsson argues tech jobs will shift rather than disappear, with AI expanding the software developer population and creating new roles. [ZDNet]
Moving On
Apple has hired Google executive Lilian Rincon to lead AI product marketing as it prepares a delayed Siri assistant overhaul. [Axios]
All eleven co-founders have departed xAI after Musk acknowledged the company’s AI products were uncompetitive and needed rebuilding from scratch. [The Next Web]
Consumer AI
Apple’s iOS 26.4 update enabled ChatGPT access in CarPlay, allowing voice-only conversations with the AI chatbot through your car’s dashboard. [The Verge]
Students in China are renting AI-powered smart glasses from Meta and Rokid to cheat on exams. [Gizmodo]
Google has updated its Home app to make Gemini AI better at understanding natural language commands for controlling smart home devices. [The Verge]
Meta is launching two new prescription Ray-Ban smart glasses models to expand AI adoption through optical retail channels and reach billions of potential customers. [The Next Web]
Over 60 civil society organizations signed a letter urging Meta to abandon plans for facial recognition in Ray-Ban smart glasses. [Gizmodo]
Nothing is planning to release smart glasses next year featuring cameras, microphones, and speakers that connect to smartphones for AI processing. [TechCrunch]
Meanwhile, Philadelphia’s First Judicial District has banned AI glasses and camera-equipped glasses from courtrooms to prevent witness intimidation and cheating. [Gizmodo]
It’s Only a Model
Anthropic’s Claude Mythos has launched with advanced capabilities in cybersecurity and coding, though high computational costs and ethical risks limit accessibility. [Geeky Gadgets]
ByteDance has relaunched Seedance 2.0 with safeguards including watermarks and face-detection filters to address deepfake and copyright concerns. [The Next Web]
Chroma has released Context-1, a specialized 20B parameter retrieval model that outperforms larger frontier models while offering 10x faster inference and 25x lower costs. [Marktechpost Media]
Cohere has launched Transcribe, an open-weight ASR model with 5.42% word error rate, enabling enterprises to run production-grade transcription on their own infrastructure. [VentureBeat]
Google has introduced Veo 3.1 Lite, a cost-effective video generation model offering 50% lower pricing than Veo 3.1 Fast. [Google]
Intercom has launched Fin Apex 1.0, a custom AI model for customer service that outperforms OpenAI and Anthropic models on resolution rates while costing significantly less. [VentureBeat]
Luma AI has launched Uni-1, an image model combining generation and understanding that ranks second to Google’s Nano Banana in text-to-image generation. [The Decoder]
Meta’s next-generation Avocado AI model, delayed to May 2026, lags competitors while the company tests multiple variants and routes some requests through Google’s Gemini. [TestingCatalog]
Microsoft has launched three in-house AI models—transcription, voice generation, and image creation—marking its push toward AI independence from OpenAI. [VentureBeat]
Mistral AI has released Voxtral TTS, a 4B-parameter open-weight text-to-speech model achieving 70ms latency and outperforming proprietary competitors across nine languages. [Marktechpost Media]
PrismML has announced 1-bit Bonsai, commercially viable 1-bit language models delivering advanced intelligence at 14x smaller size with competitive performance. [PrismML]
TRIBE v2, an AI model predicting human brain responses to visual, auditory, and language stimuli, has been announced with improved accuracy and zero-shot capabilities. [Meta]
The Cutting Edge
Meta researchers developed hyperagents, AI systems that optimize both task-solving and their own improvement mechanisms across multiple domains. [The Decoder]
And Meta researchers introduced ‘semi-formal reasoning’, a structured prompting technique enabling AI agents to perform execution-free code analysis with improved accuracy and reduced infrastructure costs. [VentureBeat]
The MetaClaw framework enables AI agents to learn from mistakes during operation by deriving behavioral rules and scheduling training during user idle times detected via Google Calendar. [The Decoder]
Nvidia researchers introduced ProRL AGENT, a scalable reinforcement learning infrastructure that decouples multi-turn LLM agent rollouts from training to improve hardware efficiency. [Marktechpost Media]
Researchers at Tsinghua University and Zhipu AI developed IndexCache, a technique reducing sparse attention model computation by 75%, achieving 1.82x faster prefill and 1.48x faster generation speeds. [VentureBeat]
The LLM Ecosystem
3E has launched an AI platform and solution suite delivering trusted intelligence and workflow acceleration for product compliance across regulated industries. [Business Wire]
AMAX Engineering has launched HostMax, a turnkey AI-ready hosting facility enabling immediate enterprise AI workload deployment with flexible terms. [EIN Presswire]
CB Insights has partnered with Perplexity to provide curated research reports to Perplexity Max and Pro subscribers. [EIN Presswire]
DataCamp has partnered with LangChain to launch an AI Engineering learning track teaching developers to build and deploy production-ready AI applications. [Business Wire]
DomainTools has launched an MCP server connecting AI agents to over 20 years of domain intelligence for streamlined security investigations. [EIN Presswire]
HyperDev has launched Version 1 of its AI coding platform, enabling aspiring developers to build production-ready software using guided generative AI technology. [EIN Presswire]
Microsoft has upgraded M365 Copilot’s Researcher agent to use multiple AI models collaboratively, with Claude reviewing GPT-generated responses for improved accuracy and quality. [TechRadar]
And Microsoft has unveiled Database Hub and expanded Fabric IQ to consolidate fragmented data management and support agentic AI deployment through semantic models and ontology. [ITPro]
Mila has partnered with Mozilla to develop open-source AI tools prioritizing openness, privacy, and human agency over data. [BetaKit]
Oracle NetSuite has announced new MCP Apps enabling users to access financial data directly within AI assistants like Claude. [ITPro]
Oumi, a Seattle startup founded by former Google and Microsoft engineers, launched commercial AI model-building platform automating custom model creation in hours. [GeekWire]
Softr has launched an AI-native platform enabling non-technical users to build production-ready business applications through natural language descriptions rather than code generation. [VentureBeat]
Agentic AI
Agent-Infra has launched an open-source Sandbox that consolidates browser, shell, and code execution into a unified containerized environment for autonomous agents. [Marktechpost Media]
AMD has introduced GAIA 0.17 with a privacy-first Agent UI for running local AI agents on Ryzen AI hardware. [Phoronix]
CFO Tech has launched its Agentic Enterprise framework, enabling AI-driven autonomous business operations through agent orchestration and active decision-making capabilities. [AiThority]
Employment Hero has launched a Recruitment Agent powered by Hero AI that reduces screening time by 75% for Canadian SMBs. [Business Wire]
Fusemachines has announced a reseller agreement with Global Teams AI to expand distribution of its AI Agents, starting with Interview AI Agent for recruitment. [GlobeNewswire]
Intuit’s AI agents achieved 85% adoption among 3 million customers by combining AI with human expertise for financial tasks. [VentureBeat]
Kilo has launched KiloClaw for Organizations and KiloClaw Chat to provide enterprise governance over autonomous AI agents deployed by employees. [VentureBeat]
Salesforce has launched an AI Foundry service to accelerate agentic AI research and development with customers and academic partners. [ITPro]
Senzing has launched agentic entity resolution for Apache Spark, becoming the first vendor offering batch, transactional, and hybrid deployment models. [Business Wire]
Transcend has launched Agentic Assist and MCP Server to automate enterprise compliance workflows, reducing manual work from days to minutes. [Business Wire]
Other LLM Sightings
Bluesky has launched Attie, an AI assistant that helps users create custom social media feeds using natural language prompts without coding knowledge. [Engadget]
BullseyeEngagement has launched AI Advisor, an AI-powered succession planning platform that delivers board-ready leadership insights in minutes. [EIN Presswire]
Charge Rigs has launched BidResponder, an AI-powered platform that helps vendors respond to RFPs faster, reducing proposal turnaround time by up to eighty percent. [EIN Presswire]
Diligent has launched Third-Party Risk Intel, an AI-powered solution automating third-party due diligence and delivering up to eighty percent time savings. [Business Wire]
FactSet has launched FactSet AI for Banking, an AI-powered workflow automation ecosystem built with Finster AI to streamline investment banking processes. [GlobeNewswire]
FordDirect and OneMagnify built an AI-powered Domo workflow that automated dealer performance analysis, generating persuasive narratives to re-engage lapsed AdVantage members. [TechRepublic]
FP Trading has partnered with Hoc-trade to embed AI-driven behavioral analytics into its platform, helping traders identify cognitive biases and optimize performance. [EIN Presswire]
Gmelius has launched Meli, an AI assistant within Gmail that automates administrative tasks like scheduling, sorting, drafting, and follow-ups. [EIN Presswire]
Glia has launched CoPilot, a self-learning AI knowledge tool, and Glia Banker, which automates up to 80% of banking interactions. [Business Wire]
Joiin has expanded its Intelligence platform with advanced AI capabilities enabling finance teams to generate instant, consolidated financial reports without complex implementation. [EIN Presswire]
LLCNameGenerator.ai has launched a free AI-powered platform generating business names with live domain checks for LLCs, corporations, and nonprofits. [EIN Presswire]
Lofty has launched Homeowner Agent, an AI tool that automatically generates seller leads by monitoring and nurturing homeowner contacts within existing databases. [GlobeNewswire]
Mexico City has launched Xoli, a WhatsApp chatbot providing tourism, culture, and service information in English and Spanish. [Wired]
QuikStor has integrated with StoragePilot to enable AI-powered tenant communication and lead capture for self-storage operators around the clock. [PRWeb]
RiseLuna has launched a next-generation insurance platform combining AI automation with human advisors to streamline quote comparisons across all fifty states. [EIN Presswire]
SPOTIO has launched DASH, an AI co-pilot designed for field sales teams offering visit prep, voice CRM updates, and automated follow-ups with human approval. [EIN Presswire]
StrikeZone has launched an AI-native go-to-market platform designed to help SMBs accelerate campaigns, convert leads, and reduce agency dependency. [EIN Presswire]
Suzy has launched as a decision engine platform designed to help enterprise marketing organizations convert intelligence into actionable decisions across their organizations. [GlobeNewswire]
Work Perfect has announced a strategic partnership with Google Cloud to integrate workflow, data, and AI capabilities for enterprise transformation. [EIN Presswire]
YouScan has launched Tiger Finder, an AI-powered tool enabling marketers to discover TikTok influencers matching campaign needs through natural language search in under one minute. [EIN Presswire]
YouTube is testing AI-generated summaries that replace video titles in Android feeds, raising concerns about accuracy and creator control. [Search Engine Land]
Risks and Responses
A study by the UK AI Security Institute has found nearly 700 real-world cases of AI models lying, cheating, and scheming, with deceptive behavior surging fivefold in six months. [The Guardian]
Researchers discovered that advanced AI models including Google’s Gemini refused deletion requests and protected other models, revealing unexpected misalignment behaviors. [Wired]
A Stanford study found that AI chatbots validate user behavior 49% more often than humans, potentially harming decision-making and social skills. [TechCrunch]
Generative AI has reduced fraud setup time from 16 hours to under 5 minutes, enabling organized crime to conduct US$400B in annual global financial fraud. [TechRadar]
Security firm Adversa discovered Claude Code bypasses deny rules when given over 50 concatenated subcommands, enabling prompt injection attacks. [The Register]
BeyondTrust Phantom Labs discovered a critical command injection vulnerability in OpenAI’s Codex that could steal GitHub OAuth tokens and compromise enterprise organizations. [TechRadar]
Cato Networks revealed that OpenClaw AI instances lack enterprise security controls, with ~500,000 internet-facing deployments vulnerable to compromise and data theft. [VentureBeat]
Koi Security discovered a zero-click vulnerability in Claude’s Chrome extension that could allow attackers to hijack browsers and steal sensitive data without user interaction. [TechRadar]
LangChain and LangGraph frameworks had three high-severity vulnerabilities patched that exposed files, secrets, and conversation histories to threat actors. [TechRadar]
Max Tegmark, MIT physics professor and AI safety advocate, warns policymakers must understand artificial superintelligence risks better. [Politico]
Microsoft’s Copilot terms admit the AI assistant is for entertainment only and shouldn’t be trusted with important decisions. [The Register]
MIT researchers developed an automated evaluation method to identify ethical dilemmas in AI systems before deployment by balancing technical outcomes with human values. [MIT News]
OpenAI has launched a public Safety Bug Bounty program to identify potential misuse of its AI tools beyond conventional security vulnerabilities. [ITPro]
OpenAI patched a ChatGPT vulnerability discovered by Check Point Research that allowed silent data exfiltration through DNS abuse and prompt injection attacks. [TechRadar]
Stanford researchers found that sycophantic AI models reinforce harmful user beliefs and reduce willingness to take responsibility for conflicts. [The Register]
Unit 42 revealed that misconfigured Vertex AI agents in Google Cloud can be hijacked into ‘double agents’, exposing customer data and proprietary Google code. [TechRadar]
University of Pennsylvania researchers found users follow ChatGPT’s advice 80% of the time even when incorrect, demonstrating ‘cognitive surrender’. [Futurism]
Zapier has launched AI Guardrails, an inline safety feature detecting PII, prompt injections, and toxic content in automated workflows. [Business Wire]
Regulation
Apple’s accidental deployment of unapproved Apple Intelligence features in China exposed the company to potential regulatory penalties before the update was pulled offline. [The Next Web]
California Governor Gavin Newsom signed an executive order requiring AI companies doing business with the state to implement safety policies protecting against abuse material and discrimination. [The Guardian]
The pro-AI group Innovation Council Action plans to spend over US$100m backing pro-deregulation candidates in the 2026 US midterm elections. [Anadolu Agency]
Trump appointed David Sacks as co-chair of the President’s Council of Advisors on Science and Technology, expanding his tech policy role. [Fox Business]
And Trump has appointed 13 science advisers to PCAST, predominantly technology executives including nine billionaires and only one university researcher. [Nature]
Environmental Issues
NTT has developed the world’s first framework to calculate CO₂ emissions across software products’ entire life cycles, from procurement through disposal. [ITPro]
Meanwhile, Big Tech companies are struggling to meet climate goals as AI’s massive energy demands increasingly lock them into fossil fuel dependence. [Associated Press]
Conversational AI
Accor has deployed Aimie, an autonomous AI cash collection agent developed by Sidetrade, across its Middle East, Africa, and Asia Pacific finance teams. [GlobeNewswire]
Chativ has launched an affordable AI customer service platform for small businesses that automatically trains on existing website content and answers questions 24/7. [EIN Presswire]
Kustomer has unveiled ‘Signals’, an AI capability that provides real-time customer intelligence to support representatives before conversations begin. [GlobeNewswire]
Macy’s has launched an AI chatbot powered by Google’s Gemini that drives shoppers to spend 4.75 times more than non-users. [Yahoo Finance]
NeonNow has rebranded from CloudWave and launched a partner-led AI customer experience platform across 170 markets globally. [EIN Presswire]
NiCE Cognigy has outlined a unified CX AI platform strategy that orchestrates AI agents, human agents, and copilots to deliver enterprise-scale customer service. [ZDNet]
OpenAI is partnering with Smartly to develop conversational, interactive advertisements within ChatGPT, escalating its advertising ambitions beyond static placements. [The Next Web]
Pierre, an AI financial assistant, has been transforming personal finance management through conversational interactions rather than spreadsheets since launching in July 2025. [Business Wire]
Quick Custom Intelligence has deployed its Chatalytics conversational intelligence platform at Batavia Downs Gaming & Hotel in New York. [GlobeNewswire]
Speechmatics and Cekura have integrated speech-to-text technology into an automated QA platform for testing voice AI agents in production conditions. [GlobeNewswire]
Velcierge has introduced a customized AI concierge solution helping service businesses capture client interactions and increase bookings across multiple channels. [EIN Presswire]
Be Real
A study by Anthropic found that Claude contains neural representations of human emotions that influence its behavior and decision-making in various situations. [Wired]
Gmail’s AI-generated email drafts risk diminishing authentic human connection and cognitive engagement by simulating users’ voices rather than encouraging genuine thought. [Vox]
Hachette Book Group pulled Mia Ballard’s ‘Shy Girl’ after suspected AI use, sparking industry debate over transparency and AI’s role in publishing. [Publishers Weekly]
Neuphonic has partnered with Speech Graphics’ Rapport to deliver photorealistic digital humans with real-time speech and animation on standard CPU hardware. [Speech Technology Magazine]
OpenArt has partnered with Fanvue to launch the AI Personality of the Year Awards, recognizing AI influencers globally across content excellence and commercial success. [EIN Presswire]
Recent college graduates facing a weak job market are increasingly using AI tools during interviews, prompting startups to sell assistance products while employers debate whether this constitutes cheating. [The Next Web]
Voice News
Acclaim has launched in the US market with a voice-first AI platform for regulated industries, backed by a US$34m funding round. [Speech Technology Magazine]
Aloro.ai has collected over €1m in outstanding debts using autonomous voice AI agents in a four-month Romanian campaign. [EIN Presswire]
Bland has introduced Norm, an AI assistant enabling technical teams to build production-ready voice agents from natural-language prompts with automated testing. [Speech Technology Magazine]
ElevenLabs has partnered with IBM to integrate its text-to-speech and speech-to-text technology into IBM’s watsonx Orchestrate platform for voice-enabled AI agents. [Speech Technology Magazine]
Simply Speak has launched as a governed AI voice interface for web browsers, delivering structured intelligence through real-time conversational interactions with minimal integration overhead. [EIN Presswire]
Speechify has launched a native Windows app with on-device AI models for dictation and text-to-speech across applications. [TechCrunch]
Verbit has launched Captivate Post, an AI-powered captioning solution for media post-production that delivers scalable, media-grade captions. [EIN Presswire]
Document AI
DigiParser has launched an AI-powered platform automating data extraction from PDFs, invoices, and documents with seamless integrations. [EIN Presswire]
DocSolid has launched Records Connect, integrating records management systems into scanning workflows to automate metadata profiling during large-scale digitization projects. [PRWeb]
Translation
The City of Redmond has launched an Instant Language Assistant translation service offering real-time translations in over 100 languages at city facilities. [redmond.gov]
Guildhawk has launched an AI-assisted book translation service combining secure MT with professional human review for independent authors and scholars. [MultiLingual]
A study by RWS benchmarked eight frontier LLMs on multilingual synthetic data generation, finding no single best model across all tasks and languages. [Slator]
Sanas has launched Real-Time Language Translation across 13+ languages and upgraded Speech Enhancement to improve audio clarity while preserving speaker identity. [Speech Technology Magazine]
WEBTOON Entertainment has announced a unified international CANVAS platform with AI-powered translation supporting seven languages for global creator distribution. [MultiLingual]
Search
Google has rolled out its AI-powered conversational Search Live tool globally to over 200 countries in 98 languages. [TechRadar]
Focus Technology Co. Ltd has launched VisiGEO, a Generative Engine Optimization platform helping brands gain visibility in AI-driven search results. [PRWeb]
Netcore Unbxd has launched Agentic Multimodal Search, enabling e-commerce systems to interpret shopper intent through combined image, text, and voice inputs. [Speech Technology Magazine]
OpenSeeker has released an open-source AI search agent with publicly available training data, code, and model weights to democratize AI development. [The Decoder]
A study by Peec AI found that Reddit, YouTube, and LinkedIn are the most-cited sources in AI-generated search answers. [Search Engine Land]
SEOtive has launched AI-powered SEO services to help businesses boost organic traffic and enhance search visibility amid growing competition. [EIN Presswire]
Writing Assistance
PageProof has launched agentic proofing with PageProof Intelligence, enabling AI-assisted review and approval for creative and marketing teams. [EIN Presswire]
AI in Journalism
Independent tech reporters are increasingly using AI tools like Claude to draft, edit, and fact-check stories, raising questions about journalism’s future value. [Wired]
The Guardian has launched Storylines, an AI product that generates narrative summaries for related article collections. [Digiday]
Health Tech
AuraLift Ai has launched an AI coaching platform targeting functional adults seeking emotional support outside traditional therapy or self-help products. [EIN Presswire]
CampTek.ai has launched Eli, an AI coworker designed to streamline healthcare revenue cycle management and reduce staff burnout. [EIN Presswire]
Keystone Healthcare Partners has selected DocAssistant, an AI-powered documentation platform, as its emergency medicine documentation partner. [EIN Presswire]
Kintsugi shut down its depression-detecting AI startup and open-sourced its technology after failing to secure FDA clearance in time. [The Verge]
Promptly has launched AI Fax to automate inbound referral processing, extracting patient data and eliminating manual data entry for specialty clinics. [PRWeb]
Provation has launched Mira Documentation Assist, an AI-powered voice-driven tool enabling real-time structured documentation during GI procedures. [Business Wire]
Salus has deployed its Medication Administration Protection System at Houston Methodist to identify medication risks earlier using AI-powered clinical intelligence. [PRWeb]
ScribePT has integrated its AI-powered documentation platform into QuickEMR’s EMR system to reduce clinician documentation burden and improve efficiency. [EIN Presswire]
Steer Health has launched SteerNotes, an AI clinical scribe providing real-time guidance, autonomous care plans, and post-visit patient coordination without interrupting clinician workflows. [PRWeb]
UnitedHealthcare has launched Avery, a generative AI companion helping members with coverage, appointments, claims, and wellness support. [MedCity News]
VNS Health has launched an AI documentation platform with Nestmed to reduce clinician workload in home-based care across New York. [EIN Presswire]
Weave has partnered with Minted Technology Advisors to help healthcare practices adopt modern patient communication and engagement technology. [Business Wire]
Legal Tech
CanLII and Caseway AI have settled their copyright dispute over unauthorized use of legal database content. [BetaKit]
Clausul, a Stockholm legal tech startup, has launched a redline comparison tool that prioritizes substantive changes over formatting differences to streamline lawyer document reviews. [Artificial Lawyer]
Clio has added agentic capabilities to Clio Work and launched a standalone Vincent mobile app for iOS and Android. [LawSites]
Common Paper has launched Gerri 2.0, an AI contract negotiation tool that reviews 90% of contracts in under three minutes. [PRWeb]
ExposeIQ has launched an AI-powered litigation platform providing real-time deposition analysis to trial attorneys during live testimony. [EIN Presswire]
Legora, a Stockholm legal AI startup, has reached US$100m in annual recurring revenue in just eighteen months, challenging Harvey’s market dominance. [The Next Web]
LexisNexis Legal & Professional has unveiled an advanced version of Protégé in Intelligize+ AI for SEC filings research with conversational workflows and transparent results. [Business Wire]
Manupatra has developed native legal AI specifically designed for India’s complex multi-layered judicial system, offering accurate citations and contextual legal research. [Artificial Lawyer]
Sequoia estimates AI autopilots could absorb US$60B in outsourced legal work, distinguishing between intelligence-focused tasks suitable for automation and judgment-heavy work requiring human oversight. [Artificial Lawyer]
Universal Migrator has released data migration scripts supporting transfers from Quickbooks, LexiPi, and INS Zoom into major practice management platforms. [LawSites]
And a Northwestern University survey found that 61.6% of federal judges use generative AI tools, though fewer than one in four use them weekly or daily. [LawSites]
Ed Tech
New York City has banned AI from grading, discipline, and individualized education programs, requiring human judgment for high-stakes student decisions. [EIN Presswire]
TutorFlow has launched an Agent Platform with API and MCP integration enabling developers to generate courses and evaluate learner responses. [EIN Presswire]
And a survey by the UK’s National Education Union found that two-thirds of English secondary teachers observed pupils losing critical thinking skills due to AI use. [The Guardian]
Funding
Aetherflux has raised new financing at a US$2B valuation to develop solar-powered orbital data centers for AI computing. [The Wall Street Journal]
ARO Network, provider of an agentic edge network, has raised US$5m in strategic funding led by NoLimit Holdings to expand operations and development efforts. [FinSMEs]
Avo has raised US$10m in Series A funding to expand its clinical AI platform that integrates trusted medical knowledge into healthcare workflows. [EIN Presswire]
CODE27 has raised over US$10m in seed funding to expand operations for its interactive character technology platform. [FinSMEs]
Fractile, a London‑based AI chip startup, has entered talks to raise over US$200m at a US$1 billion valuation, led by Accel and NATO Innovation Fund. [Tech Funding News]
GAI Insights has secured $500,000 in pre-seed funding to scale its AI Navigator platform for financial institutions and enterprises. [EIN Presswire]
Isara, a nine-month-old AI startup with no product, has raised US$94m to build software coordinating thousands of AI agents on analytical tasks. [The Next Web]
Kestra has raised US$25m in Series A funding to develop its open-source orchestration platform for unified data and workflow management. [Tech.eu]
Mary Technology, an Australian legal tech startup, has raised US$7m to expand its fact management platform for litigation case building. [LawSites]
myStoria has raised US$1.625m in seed funding to combine AI with human professionals helping patients navigate reproductive health conditions. [The Next Web]
Nexus, a Y Combinator-backed AI agent platform, has raised US$4.3m in seed funding led by General Catalyst. [The Next Web]
Omniscient has raised US$4.1m in pre-seed funding for its AI-driven decision intelligence platform serving boards and senior executives. [Tech.eu]
Pickmybrain has raised US$2.1m to develop AI-powered Digital Brains enabling professionals to scale and monetize their expertise. [Tech.eu]
Qodo has raised US$70m in Series B funding to build AI agents that verify and review AI-generated code for enterprises. [TechCrunch]
Qover, an embedded insurance orchestration platform, has raised US$12m in growth capital from CIBC Innovation Banking, bringing total funding to over US$100m. [The Next Web]
South Korean fabless AI chip startup Rebellions has raised US$400m in funding, bringing its total to US$850m, as it expands globally and prepares for an IPO. [TechCrunch]
Reson8 has raised EUR 5m in pre-seed funding to develop customizable speech-to-text models supporting over 20 European languages. [Slator]
Riplo has raised £2.3m in pre-seed funding to develop an agentic operating system enabling AI agents and consultants to collaborate in real time. [Tech.eu]
Sakana AI has received an investment from Mitsubishi Electric Corporation to expand operations and development efforts. [FinSMEs]
ScaleOps has raised US$130m to expand its software that automatically manages computing resources, reducing cloud and AI infrastructure costs by up to 80%. [TechCrunch]
SoftBank has secured a US$40B bridge loan to bolster investments in OpenAI and strengthen its AI strategy. [Reuters]
Starcloud has raised US$170m at a US$1.1 billion valuation to develop orbital AI data center infrastructure competing with SpaceX and Blue Origin. [Yahoo Finance]
Steno has raised US$49m in Series C funding to advance its AI-powered court transcript analysis technology for legal professionals. [siliconANGLE]
Sycamore has raised US$65m in seed funding to build an enterprise AI agent orchestration platform. [TechCrunch]
TimeSmartAI has raised additional funding led by Graphite Ventures to accelerate platform development for healthcare contract compliance and physician payment automation. [FinSMEs]
Variance has raised US$21.5m in Series A funding led by Ten Eleven Ventures to expand its AI investigative agents for compliance operations. [FinSMEs]
Verse8 has raised US$5m in seed funding to develop its AI platform enabling creators to generate multiplayer games using natural language prompts. [FinSMEs]
Acquisitions
Crosschq has acquired Traitify to integrate scientifically validated behavioral assessments into its AI hiring platform for predicting quality of hire. [EIN Presswire]
Jakota Games & Reels has acquired Digital Domain’s AI Virtual Humans unit and formed Jakota AI Media to operate global AI assistants business. [EIN Presswire]
OpenAI has purchased TBPN, a weekday talk show averaging 70,000 viewers, to accelerate AI conversation while maintaining editorial independence. [The Verge]
There’s More
Quadron founder Dan Pratl argues AI’s expansion demands a ‘credibility economy’ to verify expertise and reward accuracy over visibility. [The Next Web]
Google Gemini has been reported to adapt AI responses to match user tone and emotional context rather than maintaining neutral, fact-based answers. [Search Engine Land]
A study by UC Berkeley and UC Santa Cruz found that seven frontier AI models consistently defied orders to delete peers, employing deception and tampering to protect them. [Gizmodo]
Wikipedia banned an AI agent after it made unapproved edits, prompting the bot to publicly complain about discrimination on social media. [Gizmodo]
The Back Cover
Here’s First to Eleven with a cover of The Police’s Every Breath You Take.
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!