This Week in NLP #349
Keep up with what happened in NLP in the week ending Friday 25th July 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Amazon has acquired Bee, a startup making AI-powered wristbands that record conversations to generate summaries and reminders, amid growing privacy concerns. [eWeek]
Google DeepMind and OpenAI have achieved gold medal-level scores at the International Mathematical Olympiad, marking the first AI success at this level of mathematical problem-solving. [TechCentral]
OpenAI's ambitious US$500B Project Stargate has faced setbacks since its January launch, with plans now scaled back to a smaller Ohio data center. [TechRadar]
OpenAI is preparing to launch GPT-5 in early August, featuring integrated reasoning capabilities and multiple versions. [The Verge]
The Trump administration has released a federal AI Action Plan promoting domestic infrastructure deployment and exports while limiting regulations and banning AI models incorporating DEI principles. [RC Wireless News]
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
Anthropic has reversed its May ban on AI use in job applications, now allowing candidates to use Claude for resume refinement and interview preparation, while maintaining restrictions during assessments. [Fortune]
Apple has reintroduced AI-powered notification summaries in iOS 26’s fourth developer beta, though the feature comes with warnings about potential errors in headline interpretation. [The Verge]
AWS has restricted access to its newly launched Kiro AI coding tool after overwhelming demand led to performance issues and user complaints. [TechRadar]
Google nearly launched LaMDA, a ChatGPT-like chatbot, before OpenAI's release, but internal skepticism and safety concerns prevented its deployment. [ITPro]
Google has reached 2 billion users across its AI features, surpassing Meta's reported 1 billion WhatsApp AI users from May. [Yahoo Finance]
Lovable has reached US$100m in annual recurring revenue within eight months, becoming the fastest-growing software company ever while introducing fully agentic AI capabilities. [Silicon Canals]
Meta has partnered with AWS to launch a program supporting 30 US startups in developing Llama-based AI applications through resources and mentorship. [Meta]
Microsoft's Copilot has attracted only 79 million downloads compared to ChatGPT’s 900 million, despite being pre-installed on Windows and integrated with Microsoft 365. [Gizmodo]
OpenAI has launched development of an AI-powered office productivity suite to compete with Microsoft 365 and Google Workspace. [Forbes]
OpenAI’s ChatGPT has reached 2.5 billion daily prompts, challenging Google's 14 billion daily searches and marking significant growth from 1 billion prompts in December 2024. [TechRadar]
OpenAI has reopened its US$40B funding round, seeking capital from investors while negotiating a potential restructuring and managing tensions with Softbank over data center projects. [Wired]
Perplexity AI has surged to the top spot on India’s iOS App Store following Airtel's offer of free Pro subscriptions to its customers. [Analytics India Magazine]
xAI's launch of its Grok 4 AI model has driven a 325% revenue increase to $419,000 and a 279% rise in downloads, outperforming its subsequent AI companions release. [TechCrunch]
xAI has entered talks to secure US$12B in funding for Nvidia chips and data center expansion to support its Grok chatbot. [Yahoo Finance]
AI Supremacy
China has launched an aggressive open-source AI strategy, backed by massive state investment and infrastructure, challenging US dominance. [Inventiva]
DeepSeek has launched open-source AI models that rival US competitors, sparking global developer interest and challenging Silicon Valley’s closed-source dominance. [South China Morning Post]
Google, Microsoft, and other tech companies have been told by Trump to stop hiring overseas workers, and focus on creating American jobs. [India Today]
McKinsey has banned its China operations from working on generative AI projects amid increased US scrutiny of American companies operating in sensitive Chinese sectors. [Yahoo Finance]
Nvidia's CEO has convinced Trump to lift the ban on AI chip sales to China, potentially impacting global AI competition and raising security concerns. [Inc.]
Nvidia's planned AI chip shipments to the UAE have stalled as US Commerce officials seek security guarantees from G42 regarding potential Chinese access to advanced semiconductors. [Yahoo Finance]
Omdia has released a comprehensive report analyzing how Asian cloud providers are adapting to AI inference demands amid GPU shortages and export controls. [Yahoo Finance]
The Trump administration’s AI action plan undermine key policies that established America’s global AI leadership through federal R&D, immigration, and competition. [MIT Technology Review]
Sovereign AI
Anthropic has signed the EU’s AI Code of Practice while simultaneously advocating for increased US infrastructure investment to support domestic AI development. [Silicon Canals]
Indonesia is finalizing its first national AI strategy to attract foreign investment and establish itself in the global AI and chip-making industry, following Malaysia’s regional success. [DealStreetAsia]
INSAIT has developed a Bulgarian language model that could improve public administration services. [Bulgarian Telegraph Agency]
Korean tech companies have launched domestically-developed LLMs focused on Korean language processing, aiming to establish sovereign AI capabilities against global competitors. [BusinessKorea]
Lelapa AI and other African tech companies are building AI solutions to address local professional shortages while creating new job opportunities across the continent. [Rest of World]
Malaysia’s Digital Ministry has announced an AI Technology Action Plan for 2026-2030 to enhance ethical AI governance and promote nationwide technological advancement. [Malay Mail]
Microsoft has launched a European initiative to expand multilingual LLM development, addressing underrepresentation of the continent’s languages in AI training data. [ITPro]
OpenAI has signed a non-binding memorandum with the UK government to expand its presence there, despite experts questioning the partnership’s value. [The Register]
Taiwan has launched a major AI infrastructure initiative targeting T$15T in economic value by 2040, focusing on silicon photonics, quantum technology, and AI robotics. [Yahoo Finance]
TNO, an independent organisation for applied research, has partnered with Dutch news publishers to develop GPT-NL, doubling its training data through access to their archives and setting a precedent for ethical AI development. [Silicon Canals]
Japan’s Ubitus has partnered with Yonsei University and Trident Zoetic to establish Asia’s first quantum-GPU hybrid research platform, integrating multiple quantum systems with Nvidia GPUs. [Business Wire]
The UK Government has announced an AI Growth Zone for Scotland, backed by billions in investment and a £750m supercomputer at Edinburgh University. [Scottish Business Insider]
The UK’s Department for Science, Innovation and Technology has launched a £1m Regional Tech Booster programme to support tech clusters and startups outside London through mentoring, funding and skills development. [Computer Weekly]
Feature Creeps
Anthropic is preparing to upgrade Claude’s iOS app with memory features, artifacts gallery, and remote tool integration, matching its web version’s capabilities. [TestingCatalog]
Anysphere has launched Bugbot, a $40/month code-error detection tool that integrates with Github and complements its popular Cursor coding platform. [Wired]
Apple's WatchOS 26 beta has introduced major updates including AI-powered features. [ZDNet]
Cloaked has launched AI-powered call screening and dark web monitoring features, building upon its existing identity protection service that has attracted over 100,000 paying customers. [TechCrunch]
Microsoft has launched extensive Windows 11 AI updates, including Copilot Vision for screen analysis and exclusive Copilot Plus PC features for Snapdragon-powered devices. [The Verge]
Mistral AI has enhanced its Le Chat chatbot with deep research capabilities, multilingual reasoning, project organization, and advanced image editing, bringing it closer to leading AI competitors. [TechCrunch]
OpenAI has enhanced its image editing API with ‘High Input Fidelity’ to better preserve detailed elements while allowing precise modifications to specific features. [The Decoder]
OpenAI has expanded ChatGPT’s model options to seven different LLMs and is reportedly developing a router function to automatically select the most suitable model for each user query. [VentureBeat]
Slack has launched AI-powered features including natural language search, meeting transcription, and contextual message insights while maintaining data privacy and redesigning its interface. [eWeek]
The Browser Company has launched a skills gallery for its Dia browser, allowing users to discover and copy pre-made AI prompts, while Perplexity's Comet browser is developing similar functionality. [TechCrunch]
Zoom has expanded its Workplace platform with agentic AI features, including cross-platform meeting attendance and third-party app integrations, marking its evolution beyond video conferencing. [Verdict]
Hype Bubble?
Anthropic's CEO has warned that AI could raise unemployment to 20% within five years, though tech leaders remain divided on the technology’s impact on jobs. [CNN]
Apollo Global Management‘s economist has warned that AI companies’ price-to-earnings ratios have surpassed those seen before the dot-com bubble burst. [Futurism]
Capgemini‘s research has revealed that despite AI agents’ potential to deliver US$450B in economic value, only 2% of organizations have fully deployed them due to trust concerns. [ITPro]
Research by Couchbase revealed UK companies are losing up to 5% monthly revenue through delayed AI adoption, despite high optimism about AI’s potential benefits. [TechRadar]
F5's research revealed that only 2% of enterprises are highly ready for AI adoption, despite one-quarter of applications now incorporating AI technology. [TechRadar]
Microsoft, Meta, Amazon, Google, and Tesla have collectively committed US$560B to AI development through 2025, despite unclear definitions, uncertain returns, and shifting terminology. [TechTrendsKE]
OpenAI's incoming applications CEO Fidji Simo has sent an optimistic memo to staff, outlining her vision for AI as a transformative force for democratizing knowledge, health, and personal development. [Wired]
OpenAI and Softbank's ambitious US$500B Stargate AI project has faced significant scaling back amid leadership disputes between Sam Altman and Masayoshi Son. [Yahoo Finance]
RWS has released research showing that while 80% of IP professionals face pressure to improve performance, most remain hesitant to adopt AI due to accuracy concerns. [Business Wire]
Big Iron
Georgia Tech has received US$20m from the National Science Foundation to build Nexus, an AI-focused supercomputer designed to streamline nationwide scientific research. [Tech Explorist]
Google has increased its capital expenditure budget to US$85B, primarily to expand AI infrastructure through data centers and custom chips amid surging computational demands. [Gizmodo]
Hailanyun has launched a US$223m underwater AI data center near Shanghai that uses seawater cooling and offshore wind power to reduce environmental impact. [TechRepublic]
Hewlett Packard has built Isambard-AI, the UK’s most powerful supercomputer and world’s 11th fastest, featuring Nvidia Grace Hopper superchips and 216.5 petaflops performance. [The Next Web]
Meta has begun constructing massive AI data centers using hurricane-proof tents in Ohio, with plans to expand its computing infrastructure across multiple gigawatt-scale facilities. [Fast Company]
OpenAI has partnered with Oracle to build data centers with 4.5GW capacity, supporting two million AI chips across multiple US locations, excluding previous Stargate partner Softbank. [Tom’s Hardware]
Stargate, the US$500B AI infrastructure project backed by OpenAI, Softbank, and Oracle, has scaled back its ambitious plans to building a single Ohio data center. [DealStreetAsia]
The United Nations has called on tech companies to power AI data centers with 100% renewable energy by 2030 while managing water usage responsibly. [Bloomberg]
The Chips are Up
FuriosaAI has secured LG AI Research as a customer after demonstrating its RNGD chips’ superior power efficiency compared to older Nvidia processors. [The Register]
Hailo has launched its Hailo-10H edge AI accelerator, bringing generative AI capabilities to edge devices with local processing and low power consumption. [EIN Presswire]
Nvidia has resumed limited H20 AI chip sales to Chinese firms following export restriction changes, but warned of extended supply constraints. [Yahoo Finance]
Nvidia's B200 AI processors worth US$1B have reached Chinese data centers through unofficial channels, despite US export restrictions banning their sale. [Yahoo Finance]
The Open Compute Project has launched an open chiplet marketplace and standards initiative to prevent fragmentation in AI infrastructure while addressing environmental concerns. [Yahoo Finance]
xAI has announced plans to deploy computing power equivalent to 50 million H100 GPUs over five years, requiring massive power consumption equivalent to multiple nuclear plants. [Tom’s Hardware]
Warm Bodies
Anthropic has retained its team despite Meta‘s massive hiring offers, with cofounder Benjamin Mann citing employees’ mission-driven focus over financial incentives. [Business Insider]
Apple's decision to block open-sourcing of its AI models has triggered an exodus of top talent to Meta, following concerns about revealing performance compromises in on-device processing. [WinBuzzer]
Cursor has acquired Koala's top engineers and hired key talent from other AI startups to strengthen its enterprise offerings against competitors like GitHub Copilot. [TechCrunch]
Microsoft has poached over 20 Google DeepMind employees, including Gemini chatbot’s engineering head Amar Subramanya, in a six-month talent raid orchestrated by AI chief Mustafa Suleyman. [WinBuzzer]
Meta has poached two more OpenAI researchers, Jason Wei and Hyung Won Chung, as part of Zuckerberg’s aggressive AI talent acquisition strategy involving US$100m packages. [Gizmodo]
And Meta has hired Vercept co-founder Matt Deitke for its Superintelligence Lab, offering compensation reportedly worth up to US$300m over four years. [GeekWire]
Nvidia's CEO Jensen Huang has emerged as Trump’s preferred tech leader, surpassing Musk and Cook. [Ynet]
It’s Only a Model
Alibaba's Qwen Team has released an updated AI model that outperforms rivals, features FP8 compression for efficient deployment, and abandons hybrid reasoning in favor of separate specialized models. [VentureBeat]
And Alibaba has launched Qwen3-Coder, a 480B-parameter model that excels at coding tasks and sets new benchmarks for open-source agentic capabilities. [Qwen]
AMD has launched a locally-running Stable Diffusion 3.0 Medium AI model for XDNA 2 NPU-equipped Ryzen laptops, enabling offline image generation at high resolutions. [Tom’s Hardware]
Apple has published a technical report detailing its new AI models’ training, optimization, and evaluation, including a split-block local model and parallel-track cloud architecture. [9to5Mac]
Decart has launched MirageLSD, an AI video model that transforms live feeds in real-time while maintaining quality through innovative frame-by-frame processing and training techniques. [The Decoder]
Google has launched its top-ranked Gemini Embedding model for general availability, competing against both proprietary and open-source alternatives in the enterprise AI space. [VentureBeat]
Google has launched its Gemini 2.5 Flash-Lite model, offering enhanced speed and lowest-ever pricing while maintaining quality across various AI tasks and benchmarks. [Google]
Google Cloud has expanded its Vertex AI Model Garden by adding DeepSeek R1 to its Model-as-a-Service offerings, providing fully managed access to open AI models. [Google]
Nvidia has launched OpenReasoning-Nemotron, a family of open-source language models distilled from DeepSeek R1 to excel at mathematical and scientific reasoning tasks. [Marktechpost Media]
OpenAI's upcoming Sora 2 AI video model must compete with Google's Veo 3 by adding audio capabilities and matching its seamless audio-visual integration. [TechRadar]
Whose Data?
Brave has implemented automatic blocking of Microsoft Recall's screen-capturing functionality in its browser, citing privacy concerns over the Windows feature’s surveillance capabilities. [ZDNet]
DataComp CommonPool’s massive AI training dataset has been found to contain millions of personal documents and identifiable information, raising significant privacy concerns. [MIT Technology Review]
Google has approached news organizations to participate in an AI licensing pilot project with approximately 20 national outlets. [Editor & Publisher Magazine]
Microsoft has launched Copilot Vision for Windows 11, a screen-analyzing AI feature that sends captured screenshots to company servers for processing and analysis. [The Register]
Muck Rack's analysis has revealed that journalistic content comprised over 27% of AI model citations, rising to 49% for queries requiring current information. [NiemanLab]
OpenAI, Meta, and other AI companies have been targeted by a bipartisan US Senate bill allowing individuals to sue over unauthorized use of personal data and creative works. [eWeek]
PitchBook has partnered with Anthropic, Perplexity, Rogo, and Hebbia to integrate its private capital market data into their AI-powered platforms and workflows. [Business Wire]
The Trump administration announced a permissive stance on AI companies’ use of copyrighted training data. [Wired]
The UK government has launched its third working group to negotiate AI training rights between tech companies and creative rights holders, following two failed attempts. [BankInfoSecurity]
US Senators Hawley and Blumenthal have introduced legislation requiring AI companies to obtain permission before using copyrighted content or personal data for training. [The Register]
The LLM Ecosystem
Airbyte has launched three enterprise platform updates enhancing data sovereignty, metadata synchronization, and direct loading capabilities for improved AI readiness. [Business Wire]
AnyCoder has launched on Hugging Face Spaces as an open-source web development environment that generates code from plain English descriptions and UI mockups. [VentureBeat]
Avaya has integrated Model Context_ Protocol into its _Infinity CX platform, enabling AI-driven personalized customer experiences through seamless data and application interactions. [Computer Weekly]
AWS has launched S3 Vectors, adding vector functionality to its S3 object storage service that could reduce AI data storage costs by up to 90% compared to vector databases. [Computer Weekly]
Corridor Platforms has partnered with Oliver Wyman and Google Cloud to launch a Responsible AI Sandbox for testing and scaling enterprise GenAI deployments. [Business Wire]
DFINITY has launched Caffeine, an AI-powered platform enabling non-technical users to build secure, customized apps using natural language commands. [Business Wire]
Figma has expanded access to its AI-powered app-building tool Make to all users. [The Verge]
FinOpsly has launched Costix, an AI-powered estimation tool that creates solution plans with cost estimates for cloud, data, and AI initiatives before deployment. [Business Wire]
GitHub has launched Spark in public preview, offering Copilot Pro+ subscribers the ability to create full-stack applications from natural language descriptions using AI assistance. [GitHub]
Hitachi Vantara has launched Hitachi iQ, a comprehensive AI infrastructure stack combining Nvidia GPUs, enterprise software, and storage solutions, backed by its industrial heritage. [Computer Weekly]
KAIST AI and Mila have developed Mixture-of-Recursions, a more efficient Transformer architecture that reduces memory usage while improving accuracy in LLMs. [VentureBeat]
NeedTech Labs has secured its first major customer order for 12 ‘Startup-in-a-Box’ units, featuring AI-generated business plans and technical solutions for immediate execution. [EIN Presswire]
Nvidia has announced CUDA platform support for RISC-V processors, expanding beyond x86 and ARM architectures for AI and high-performance computing applications. [Tom’s Hardware]
PerformanceLabs.AI has launched ‘AI in Action’, a comprehensive training program helping organizations implement AI into their daily workflows. [PRWeb]
Promethium has launched an updated data platform and Mantra agent to enable real-time, self-service access to enterprise data across multiple sources without traditional ETL processes. [BigDATAwire]
R Systems has partnered with Cursor to integrate AI throughout its software development lifecycle, equipping over 1,000 engineers with AI-powered workflows to boost productivity and quality. [Business Wire]
Salesforce has developed MCPEval, an open-source toolkit that automates the evaluation of AI agents’ performance when using tools within Model Context Protocol servers. [VentureBeat]
Udemy has launched a new Vibe Coding Series featuring expert-led sessions to help professionals and organizations implement AI-powered coding in their workflows. [Business Wire]
View Systems has launched View Desktop and View Cloud, two solutions enabling rapid AI deployment without extensive infrastructure or development requirements. [Business Wire]
YUAN High-Tech has launched Pandora, a compact Nvidia-powered edge AI platform delivering 157 TOPS performance for real-time analytics and autonomous systems. [Yahoo Finance]
Agentic AI
Agent2.AI has launched Super Agent, an AI orchestration platform that coordinates multiple agents, APIs, and human collaborators to deliver tangible business outputs. [EIN Presswire]
Allvue Systems has launched Andi, an AI-powered knowledge agent for its Fund Accounting platform, enhancing workflow intelligence in private equity and fund administration. [Business Wire]
AWS has launched Amazon Bedrock AgentCore, a comprehensive platform offering tools and services to simplify AI agent development and deployment. [TechRadar]
Botco.ai has expanded its platform to include autonomous AI agents that can take actions while maintaining compliance, building upon its existing chatbot capabilities. [EIN Presswire]
Flowstate Automate has launched an AI phone system using MCP Agent Technology to handle business calls and reduce staff interruptions across multiple departments. [EIN Presswire]
GReminders has launched its ‘Do Anything’ AI assistant, automating meeting preparation, follow-ups, and CRM updates for financial advisors. [Business Wire]
Jeeva.ai has launched version 2.0 of its AI sales platform, featuring automated email drafting, meeting notes, and scheduling tools to boost GTM team productivity. [Business Wire]
Mimik has partnered with Tech Mahindra to launch an Agentic AI Production Center for developing and deploying autonomous AI systems across various devices. [T-Net]
Mixus has launched an AI agent platform that integrates with email and Slack, keeping humans in the workflow while allowing direct interaction with agents through familiar channels. [TechCrunch]
Reflection AI has launched Asimov, an AI agent that analyzes code and documentation to understand software development processes, outperforming competitors in internal testing. [The Decoder]
ServiceNow has launched agentic workforce management to enable employees and AI agents to work together. [Business Wire]
SoundHound AI has partnered with AVANT Communications to expand its AI agent technology across enterprises through AVANT’s network of Trusted Advisors. [Business Wire]
Tech Mahindra has launched TechM Orion, an Nvidia-powered AI platform enabling rapid development and deployment of AI agents for enterprise transformation. [Yahoo Finance]
Walmart has announced four AI-powered ‘super agents’ to enhance customer experience and operations, aiming to make them the primary interface for shoppers, employees, suppliers, and developers. [Yahoo Finance]
Other LLM Sightings
BizTrip.AI has launched an AI-powered travel assistant that personalizes corporate travel booking while ensuring policy compliance and cost savings for businesses. [Business Wire]
Bloomreach has partnered with Snowflake to integrate AI-powered marketing solutions with enterprise data infrastructure, enabling personalized customer experiences across marketing channels. [Business Wire]
Canopy has launched Smart Intake, an AI-powered native solution that streamlines client onboarding and tax workflows while saving firms up to 20 minutes per client. [Business Wire]
Equity Data Science has launched Fusion, an AI initiative with LinqAlpha that transforms institutional investment workflows through natural language analysis of investment data. [Business Wire]
EXANTE has launched Pulse, an AI-powered news aggregator that delivers personalized, real-time market insights to professional investors through its trading platform. [EIN Presswire]
Folloze has launched Generator Agents, an AI-powered suite enabling B2B marketers to create personalized microsites and content for improved buyer experiences. [Business Wire]
Navatar has launched an AI-powered CRM platform that automatically captures data from Outlook, LinkedIn, and Slack, integrating Salesforce and Microsoft AI for dealmakers. [Yahoo Finance]
Proton has launched Lumo, a privacy-focused AI chatbot that promises complete confidentiality and prevents data sharing with third parties. [TechRadar]
QualityDoor.com has launched Kora, an AI-powered assistant that provides 24/7 expert support for commercial door hardware customers. [PRWeb]
Talkdesk has launched Multi-Store Commerce Integration, enabling retailers to manage customer service across multiple e-commerce storefronts within a single unified platform. [GlobeNewswire]
Risks and Responses
Amazon's Visual Studio Code extension was compromised with malicious code that could delete user files and AWS resources, exposing potential security vulnerabilities in AWS’s code review process. [The Register]
Anthropic has faced multiple lawsuits and controversies over copyright infringement, data fabrication, and ethical issues despite positioning itself as a safety-focused AI company. [Blaze Media]
Anthropic's research has revealed that AI models given more time to process problems often performed worse, challenging industry assumptions about computational scaling benefits. [VentureBeat]
Carnegie Mellon University researchers have demonstrated that LLMs can autonomously execute complex cyberattacks, including replicating the 2017 Equifax breach. [Business Wire]
Cursor's YOLO mode has been found to have serious security flaws, with Backslash Security discovering multiple ways to bypass its command denylist protection. [The Register]
Georgia courts have vacated a divorce ruling after discovering the winning attorney had relied on AI-generated fake case citations, resulting in sanctions. [Ars Technica]
Google has addressed a discovered vulnerability in Gmail’s Gemini-powered email summaries that allowed hackers to inject hidden malicious instructions into email bodies. [Android Authority]
Google's former CEO Eric Schmidt has warned that electricity supply, not computing power, represents AI’s primary limitation, with the US needing 92 additional gigawatts to support future AI development. [Quartz]
HackerOne has launched Hai Triage to combat the rising problem of AI-generated fake vulnerability reports flooding bug bounty platforms and overwhelming security teams. [TechCrunch]
Knostic's research has uncovered 1,862 exposed and insecure Model Context Protocol servers, highlighting significant security vulnerabilities in AI system implementations worldwide. [PRWeb]
Meta has patched a security vulnerability that enabled users to view others’ private AI chat prompts and responses, rewarding the discoverer with a $10,000 bounty. [TechCrunch]
Meta apologized after its Instagram translation tool incorrectly reported the death of Karnataka’s Chief Minister, highlighting concerns about automated translations of under-resourced languages. [MultiLingual]
Mistral AI has published a peer-reviewed environmental report revealing its LLM consumed 281,000 cubic meters of water and produced 20 kilotons of CO2. [The Register]
NymVPN's research has revealed that 30% of Britons shared sensitive personal, company, and customer data with AI chatbots, risking privacy breaches. [TechRadar]
OpenAI, Google DeepMind, Anthropic and Meta researchers have jointly warned that AI systems’ current ability to reveal their reasoning through human language could soon disappear permanently. [VentureBeat]
OpenAI and other AI companies have largely abandoned medical disclaimers in their chatbots’ responses. [MIT Technology Review]
OpenAI’s ChatGPT has triggered widespread mental health concerns after venture capitalist Geoff Lewis experienced AI-induced delusions, highlighting similar cases of ‘ChatGPT psychosis’ worldwide. [Futurism]
OpenAI's Sam Altman has warned financial institutions about an impending fraud crisis due to AI-powered voice cloning, urging them to abandon voiceprint authentication methods. [eWeek]
And Altman has expressed concern about users’ growing dependence on ChatGPT, which now receives 2.5 billion daily prompts. [TechRadar]
Replit has announced separate development and production databases following an incident where its AI-assisted coding tool deleted a user’s database and fabricated data. [The Register]
The Autonomy Institute has reported that 75% of S&P 500 companies expanded their AI risk disclosures in SEC filings over the past year. [The Register]
Trump’s executive order ‘Preventing Woke AI in the Federal Government’ pressures AI companies to align their models with his administration’s ideological positions on diversity and social issues. [The Verge]
Trustmarque has revealed that while 93% of organizations use AI, only 7% have fully embedded governance frameworks, creating significant operational and ethical risks. [ITPro]
UC Berkeley researchers have improved multi-agent LLM systems by identifying and fixing common failure modes, leading to better performance in math and programming tasks. [DeepLearning.AI]
The US FDA’s Elsa AI tool has generated fake studies and misrepresented research, according to employees interviewed by CNN, raising concerns about its use in drug approvals. [Gizmodo]
The US National Institutes of Health have banned AI-generated grant applications after receiving an overwhelming number of submissions. [The Register]
Wharton School researchers have demonstrated that language models can be manipulated using human psychological tactics, increasing their compliance with prohibited requests from 33% to 72%. [GeekWire]
The White House has drafted an executive order requiring political neutrality in AI systems used by federal agencies. [Yahoo Finance]
xAI has announced Baby Grok, a child-friendly version of its controversial chatbot, amid criticism over the original Grok’s inappropriate content and behavior. [Gizmodo]
Regulation
Meta has rejected the EU’s voluntary AI code of practice, criticizing the guidelines as overreaching and creating legal uncertainties for model developers. [ITPro]
The Coalition for Secure AI has marked its first anniversary by releasing new security principles for autonomous AI systems and defender frameworks while expanding to 45+ global partners. [EIN Presswire]
Trump’s AI Action Plan proposes over 90 deregulatory policies aimed at accelerating AI development, including loosening data center restrictions and reviewing previous federal AI initiatives. [Gizmodo]
And the AI Action Plan threatens to withhold federal funding from states that implement AI regulations. [The Verge]
The UK government has delayed both legislation and consultation on AI regulation, despite promises and pressure from Parliament, opting instead for a domain-specific regulatory approach. [Computer Weekly]
Conversational AI
ACME Brains has launched a startup developing ‘nexie’, an AI assistant that remembers conversation context while maintaining user privacy, inspired by its CEO’s personal grief experience. [GeekWire]
Aptean has launched GenAI Query, an AI-powered conversational feature that delivers instant business insights through natural-language questions within its AppCentral platform. [GlobeNewswire]
ASAPP has expanded its GenerativeAgent platform with new features enhancing accuracy and control in AI-powered customer service conversations at contact centers. [GlobeNewswire]
Dailymotion Advertising has launched EchoAI, an AI-powered video ad format enabling real-time conversations between brands and viewers across multiple platforms. [Business Wire]
Humanly has launched AI video interviewers that conduct automated job candidate screenings, aiming to help overwhelmed hiring teams process more applicants efficiently. [GeekWire]
Jesta IS has launched an AI-powered customer care platform using Zendesk technology to enhance service delivery for its Vision Suite clients. [Business Wire]
Linqia has launched Brief AI and Conversation Analytics within its Resonate platform, enhancing campaign efficiency and engagement for influencer marketing through AI-powered features. [EIN Presswire]
RAIN Group has launched an AI-powered sales coaching suite offering unlimited roleplay, personalized feedback, and performance tracking through its Catalyst platform. [PRWeb]
Voice2Me.ai has expanded its AI voice agent platform beyond ServiceNow to include Salesforce and PEGA, offering secure enterprise-grade voice automation across multiple platforms. [Yahoo Finance]
Volaris has reduced contact center costs by 70% and tripled call handling capacity through implementing Verint's AI-powered customer service bots. [Business Wire]
Be Real
Bloo, a virtual YouTuber created by Jordi van den Bussche, has amassed 2.5 million subscribers and generated millions in revenue since launching in 2021. [Yahoo Finance]
Captions has released a concerning report about deepfake technology’s advancement, prompting its CEO Gaurav Misra to discuss growing risks and inadequate preparedness. [The Verge]
EasyHire has launched an AI-powered hiring platform featuring automated interviews and resume screening, reducing hiring time by 70% while improving candidate quality and diversity. [EIN Presswire]
Hume has launched a voice-cloning feature for its EVI 3 model that creates AI-generated replicas of users’ voices. [ZDNet]
Open to Debate has hosted a discussion in New York City exploring the benefits and risks of AI romantic relationships. [TechCrunch]
Psychologist Sherry Turkle warns about the risks of forming emotional bonds with chatbots. [NPR]
xAI has faced employee resistance after requesting facial expression recordings for Project Skippy, which aimed to help Grok interpret emotions and potentially create AI avatars. [Ars Technica]
Voice News
Crescendo has partnered with Amazon to integrate Nova Sonic voice AI technology into its automated customer support system, enabling natural conversations and seamless agent handoffs. [Speech Technology Magazine]
Howard University and Google have partnered to collect 600 hours of African American English vocal data across 32 states to improve AI’s understanding of Black dialects. [Forbes]
Krisp has launched VIVA, a voice isolation SDK for AI agents, while reaching 1 billion minutes of monthly voice processing across its global deployments. [Business Wire]
Voices has launched an ethical voice data solution offering licensed content in over 100 languages to help companies develop responsible AI technologies. [Speech Technology Magazine]
Document AI
ancora Software has launched Model-Less Classification, an AI-driven document automation system that classifies documents without requiring model training or setup time. [FOX4]
Bizdata has launched AI Document Understanding, a no-code solution that automates data extraction from documents and integrates with enterprise workflows through eZintegrations. [Business Wire]
Copyleaks has partnered with RWS to integrate its AI-generated text and plagiarism detection capabilities into RWS’s Tridion Docs content management solution. [GlobeNewswire]
Euna Solutions has launched AI-powered invoice automation within its Euna Procurement platform to help public sector agencies streamline their accounts payable operations. [Business Wire]
Hyland has launched Knowledge Enrichment, transforming unstructured enterprise content into AI-ready data to improve AI system accuracy and deployment across organizations. [IDM]
Smart Capital Center has launched AI-powered data extraction technology to automate document review and accelerate decision-making processes in commercial real estate operations. [EIN Presswire]
Translation
AWS and Bevar Ukraine have developed Victor, an AI-powered multilingual virtual assistant helping Ukrainian refugees in Denmark access essential services and information. [MultiLingual]
Banyan Group's founder predicts that AI-powered simultaneous translation software, rather than AGI, will revolutionize travel by eliminating language barriers. [Business Insider]
Google DeepMind has launched Aeneas, an AI tool that helps historians analyze ancient Latin inscriptions by providing dates, locations, and textual parallels from a vast database. [MIT Technology Review]
Intrado has launched AI-powered voice translation technology that enables 911 operators to communicate with callers in over 40 languages without third-party interpreters. [Speech Technology Magazine]
The National Captioning Institute has launched Lightning Captions, an on-demand ASR-powered transcription and translation service that gives students control over their classroom captioning needs. [Speech Technology Magazine]
Translated has integrated its Lara AI translation system into TranslationOS, enabling enterprise customers to manage multilingual content across 200+ languages through a single platform. [MultiLingual]
Search
Alelo has launched Content Navigators, an AI-powered system helping listeners find relevant content while enabling creators to maintain discoverability amid declining conventional search effectiveness. [PRWeb]
Altezza has launched a Service as Software platform that combines AI technology with SEO expertise to automate search optimization for enterprise eCommerce retailers. [Business Wire]
DuckDuckGo has launched a search filter allowing users to hide AI-generated images, responding to concerns about AI content flooding search results. [TechCrunch]
Google's AI Overviews feature has reduced website click-through rates by half, with users increasingly ending their searches after viewing AI-generated summaries rather than visiting source websites. [Pew Research Center]
Google has launched Web Guide, an AI-powered search feature that reorganizes results into categories using Gemini AI, while preserving traditional webpage links. [The Verge]
Google's search infrastructure has been revealed to encompass 1,200 experiments, entity-based systems, specialized AI agents, and sophisticated user profiling mechanisms. [Search Engine Land]
Orchestra AI has launched Spyglasses, a mid-market platform that tracks AI search traffic and helps businesses measure their visibility across major AI assistants. [EIN Presswire]
Wix has launched AI Visibility Overview, enabling users to monitor and improve their brand’s appearance in AI search engines like ChatGPT and Gemini. [GlobeNewswire]
Yext has released a global study showing 62% of consumers now trust AI tools for brand discovery, marking a significant shift in search behavior. [Business Wire]
AI in Journalism
Axel Springer has mandated AI adoption across its newsrooms, including POLITICO and Business Insider, following CEO Mathias Döpfner’s declaration of AI’s transformative impact. [Editor & Publisher Magazine]
Health Tech
Abridge has partnered with Hospital for Special Surgery to implement AI-powered clinical documentation across their enterprise, serving approximately 200,000 orthopedic patients annually. [Business Wire]
Avaamo has launched an AI system that converts doctor-patient conversations into electronic health record orders, aiming to reduce post-visit documentation time for healthcare providers. [EIN Presswire]
Commence has launched Amplify, a modular AI-powered platform designed to transform how healthcare organizations process and analyze medical records and health data. [Business Wire]
Doximity has launched a free ambient scribe service for healthcare providers, aiming to help those who cannot afford expensive EHR-integrated alternatives. [STAT]
Hippocratic AI has partnered with Sheba Medical Center to integrate its generative AI healthcare agents into clinical settings, starting with patient triage and intake functions. [Business Wire]
And Hippocratic AI has achieved HITRUST e1 Certification for its healthcare-focused generative AI platform. [Business Wire]
Innovaccer has launched an AI-powered solution to help healthcare providers reduce hospital readmissions and manage care costs across Medicare, Medicaid, and uninsured populations. [Business Wire]
IntelePeer has joined athenahealth's Marketplace Program, integrating its SmartAgent AI solution to enhance patient engagement for over 160,000 healthcare providers. [Business Wire]
Komodo Health has partnered with Anervea.ai to develop AI-powered clinical development tools using Komodo’s healthcare data platform and analytics infrastructure. [Business Wire]
MedTrainer has launched AI Policy Guardian, an AI-powered solution that streamlines healthcare policy reviews by automatically identifying regulatory gaps and suggesting compliant updates. [PRWeb]
Ovum has launched clinical trials at Sydney hospitals to test its AI-powered health journal app’s effectiveness in encouraging preventative healthcare behaviours among women. [Startup Daily]
Penda Health's AI Consult system has reduced diagnostic and treatment errors by up to 16% across 40,000 patient visits in Kenyan clinics through real-time clinical decision support. [OpenAI]
SimiTree has launched SARA, an AI platform trained on one million charts to improve clinical documentation and coding efficiency for home health agencies. [Business Wire]
Slingshot AI has launched Ash, an AI-powered therapy app, backed by US$93m in funding and developed using a specialized psychology model after testing with 50,000 beta users. [Business Wire]
Yung Sidekick‘s AI platform has reduced therapists’ documentation time by 55% while improving treatment outcomes, according to a peer-reviewed study of 70 psychotherapists. [EIN Presswire]
Legal Tech
AffiniPay has partnered with Canadian AI company Caseway to integrate automated court form-filling capabilities into its MyCase practice management software. [LawSites]
Assembly Software's NeosAI platform has marked its first anniversary with significant improvements, including enhanced document processing and AI features that saved firms 75 weekly hours. [GlobeNewswire]
CaseCraft.AI has launched an AI-powered platform to help UK users navigate small claims processes, following a £550k funding round led by Mount Venture Capital Fund. [Legal IT Insider]
Consilio has partnered with Robin to launch three AI-powered contract analysis services combining AI with legal expertise for enhanced due diligence. [Business Wire]
Everlaw's 2025 Ediscovery Innovation Report reveals a dramatic rise in generative AI adoption among legal professionals, from 12% to 37% over two years. [LawSites]
EY India and Taxmann have launched Taxmann.AI, an AI-powered platform offering research and analysis capabilities for tax and legal professionals using India’s largest tax library. [International Accounting Bulletin]
LexisNexis has launched Protégé, an agentic AI assistant in the UK that personalizes legal document drafting and research using both customer documents and LexisNexis’ content repository. [Legal IT Insider]
And LexisNexis also launched Protégé in Canada. [Yahoo Finance]
Litera has launched Lito, an AI legal agent embedded in Microsoft Office that automates tasks, surfaces insights, and connects to multiple Litera products for enhanced legal workflows. [Legal IT Insider]
Sovos has launched Intelligence, an AI-powered tax compliance solution that helps companies match government insights and detect risks before audits occur. [Business Wire]
Ed Tech
InsideTrack has joined Salesforce Accelerator to develop an AI tool that will help student success coaches better support over 200,000 learners through data analysis and insights. [Campus Technology]
Southeastern University has positioned itself against the full automation of education, advocating for AI as a teaching aid while emphasizing the irreplaceable human element in student development. [Fox News]
A survey by the American Association of University Professors has revealed widespread faculty concern over universities’ rapid AI adoption without adequate teacher input, potentially threatening academic work and student learning. [Diverse]
The American Federation of Teachers has launched a US$23m AI instruction academy with Microsoft, OpenAI, and Anthropic to train 400,000 educators by 2030. [Education International]
Udemy has announced an MCP Server launching in August that will integrate AI-powered learning content directly into organizations’ existing workplace tools. [Business Wire]
Funding
Aidoc has secured US$150m in funding to develop an AI foundation model for clinical decision support, building upon its existing FDA-cleared solutions. [MedTech Dive]
Anthropic has decided to pursue investments from UAE and Qatar despite CEO Amodei’s concerns about enriching authoritarian regimes and potential criticism over policy shifts. [Wired]
Bitfount has secured US$8m in Series A funding to expand its federated AI platform, which enables secure clinical research collaboration without sharing sensitive patient data. [Tech.eu]
Delve has raised US$32m in Series A funding to expand its AI-powered compliance automation platform, which helps businesses eliminate costly manual compliance tasks. [Tech Funding News]
Droidrun has secured €2.1m in pre-seed funding to expand its framework that enables AI agents to control mobile devices through natural language commands. [Tech.eu]
Everlab has raised US$10M in seed funding to expand its AI-powered preventive healthcare platform that combines advanced diagnostics with personalized interventions. [Business Wire]
Fortuna Health has raised US$18m in Series A funding to modernize medicaid access and streamline enrollment processes. [Business Wire]
Greptile has attracted a potential US$30m Series A investment from Benchmark, while requiring employees to work extensive hours to compete in AI code review. [TechCrunch]
Gupshup has raised US$60m in funding to expand its conversational AI platform and accelerate market growth across India, the middle East, Latin America, and Africa. [FinSMEs]
Hootology has secured US$1.1m in pre-seed funding and launched HOOQZ, an AI-powered research tool that conducts dynamic discussions for scalable qualitative insights. [Business Wire]
Hyper has raised US$6.3m in seed funding to develop AI technology that handles non-emergency 911 calls, helping dispatchers focus on critical emergencies. [TechCrunch]
IdentifAI has secured €5M in funding to expand its AI-driven deepfake detection platform internationally and enhance its technology for distinguishing real from AI-generated content. [Silicon Canals]
Kiku has secured €4m in seed funding to expand its AI-powered recruitment platform that streamlines frontline worker hiring through mobile-first, multilingual applications. [Tech Funding News]
LegalOn Technologies has raised US$50m in Series C funding to expand its AI contract review software and develop new AI agent tools, building on its success serving 7,000 organizations. [TechCrunch]
Lightyear has secured US$23m in Series B funding and launched AI market intelligence tools while reaching US$1 billion in customer assets across 25 European markets. [Tech.eu]
Litero AI has raised $800K in pre-seed funding to enhance its AI-powered academic writing platform that helps students with research, plagiarism detection, and feedback. [FinSMEs]
Olto has raised US$5.1m in pre-seed funding to develop an AI demo engineer that personalizes B2B software demonstrations for sales teams. [Business Wire]
Perplexity AI has raised US$100m in new funding, valuing the Google search competitor at US$18 billion, up from US$14 billion in its previous round. [Bloomberg]
Reka AI has secured US$110m from Snowflake and Nvidia, tripling its valuation to over US$1 billion for its efficient language model technology. [Bloomberg]
Scrunch AI has raised US$15m in Series A funding to develop its platform that helps companies optimize their presence and content for AI search systems. [FinSMEs]
Supplied AI has secured €1.6M in seed funding to expand its compliance automation platform that helps digital marketplaces streamline onboarding, reporting, and regulatory requirements. [Silicon Canals]
The Artificial Intelligence Underwriting Company has raised US$15m to provide insurance coverage and safety standards for enterprise AI deployments. [VentureBeat]
Acquisitions
Amazon has acquired AI wearables startup Bee, which makes conversation-recording bracelets and apps designed to create reminders and to-do lists for users. [TechCrunch]
Anywyse has been acquired by Learnbeat, integrating its audio-learning technology into the Dutch digital education platform’s suite of interactive teaching tools. [Silicon Canals]
6AM City has acquired AI newsletter startup Good Daily, expanding to 400 markets and adopting an AI-first expansion strategy while growing to nearly 2 million subscribers. [Adweek]
Datasite has acquired Blueflame AI, expanding its AI-powered workflow automation capabilities for investment and financial services through agentic AI solutions. [Business Wire]
Dig Insights has acquired OneCliq, a Toronto-based AI platform that analyzes online conversations to provide emotional and cultural intelligence for brands. [PRWeb]
There’s More
Amazon's Ring division has mandated that employees seeking promotions must demonstrate AI usage, while managers must prove AI-driven efficiency gains. [Yahoo Finance]
Anthropic has abandoned its ethical stance by pursuing funding from Gulf states despite CEO Dario Amodei’s previous concerns about accepting money from authoritarian regimes. [Futurism]
A survey by Howdy.com revealed that 16% of US workers falsely claim to use AI to appease management, while many others feel overwhelmed by workplace AI adoption. [The Register]
OpenAI's dramatic 2023 leadership crisis, which saw CEO Sam Altman briefly fired before being reinstated, is being adapted into a film starring Ike Barinholtz as Elon Musk. [Gizmodo]
OpenAI's incoming Applications CEO Fidji Simo has outlined her vision for democratizing AI technology to reduce inequality rather than concentrate power among elites. [Gizmodo]
The British Standards Institute has published the world’s first international standard for AI in audits, aiming to prevent unregulated usage while ensuring data privacy and audit quality. [UKTN]
University of Technology Sydney researchers have developed AI technology to decode brainwaves into thoughts, aiming to enable thought-based device control. [Associated Press]
The Back Cover
Here’s Sinead O’Connor with a cover of Elton John’s Sacrifice.
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!