This Week in NLP #350
Keep up with what happened in NLP in the week ending Friday 1st August 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Anthropic has overtaken OpenAI in enterprise AI model usage, holding 32% market share compared to OpenAI’s 25%. [TechCrunch]
China has released a global AI action plan and proposed establishing an international cooperation organization, contrasting with the US’s more unilateral approach to AI development. [CNBC]
Microsoft became the second publicly traded company to reach a US$4T market cap, boosted by strong AI partnerships and cloud services revenue. [Yahoo Finance]
And Microsoft has launched Copilot Mode for Edge browser, offering AI-assisted web browsing with features like research assistance, task automation, and voice input capabilities. [TechCrunch]
Nvidia has ordered 300,000 additional H20 AI GPUs from TSMC to meet Chinese demand, bringing its China-bound inventory close to 1 million units. [Yahoo Finance]
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 implemented weekly rate limits for Claude to curb excessive usage and account sharing. [TechCrunch]
Apple has signalled its commitment to AI by increasing investments, acquiring companies, and developing over 20 AI features across its platforms. [TechCrunch]
Google has launched Opal, an experimental AI tool that enables users to create apps through natural language prompts and visual aids without coding knowledge. [ZDNet]
Google has reported significant growth in AI adoption across its products, with AI Overviews reaching 2 billion monthly users and Gemini attracting 450 million users. [TechCrunch]
Google's AI Overviews has reduced website click-throughs while increasing its own ad revenue through strategically placed advertisements within AI-generated summaries. [The Register]
Meta has shifted its AI strategy from primarily open-source development to a mixed approach, following massive investments and a new focus on ‘personal superintelligence’ through proprietary devices. [TechCrunch]
The newly formed Meta Superintelligence Labs is focused on building AI models that can self-improve. [Wired]
Microsoft has been testing a new smart mode for Copilot that appears designed to accommodate GPT-5’s upcoming release in early August. [The Verge]
Microsoft has entered advanced negotiations with OpenAI to secure long-term access to the startup’s technology, even after potential AGI development. [Yahoo Finance]
NTT Data has partnered with Mistral AI to develop and deploy secure enterprise-grade AI solutions for regulated sectors worldwide. [Business Wire]
OpenAI has announced plans to launch GPT-5 in early August, combining reasoning and GPT-series models while offering tiered access levels for different subscribers. [ZDNet]
OpenAI's Sam Altman has expressed fear about GPT-5’s capabilities during testing. [TechRadar]
And OpenAI doubled its annual revenue to US$12B in the first seven months of 2024. [DealStreetAsia]
Perplexity AI has experienced explosive growth in India through strategic partnerships and marketing, reaching 3.7m monthly active users and surpassing US$100m in annual revenue. [Analytics India Magazine]
xAI has launched ‘Imagine’, a text-to-image/video AI tool, backed by 230,000 Nvidia GPUs, to compete with Meta and OpenAI. [Social Media Today]
AI Supremacy
Chinese and Western tech companies gathered in Shanghai for the World AI Conference, where Premier Li Qiang addressed China’s AI ambitions amid US sanctions. [Yahoo Finance]
China has proposed establishing a global AI cooperation organization amid ongoing US-China tensions over AI development and chip export controls. [Gizmodo]
Chinese AI companies have formed two industry alliances to reduce foreign tech dependence. [Profit]
China’s AI hardware development has closed performance gaps with Western competitors but remains hindered by manufacturing limitations and software ecosystem weaknesses, keeping it dependent on Nvidia chips. [Epoch AI]
Nvidia's approval to sell H20 AI chips in China has prompted 20 national security experts to urge the Commerce Department to reverse its decision. [TechCrunch]
Nvidia has denied having backdoors in its H20 AI chip after China’s cybersecurity regulator raised security concerns about potential remote access risks. [Reuters]
StepFun has formed a Model-Chip Ecosystem Innovation Alliance with Chinese AI developers and chipmakers to build a domestic supply chain and reduce foreign dependencies. [Asia Financial]
Trump’s AI Action Plan emphasizes deregulation and global AI dominance while focusing on worker upskilling over protections and relaxing environmental standards for data centers. [ZDNet]
Trump’s ‘Big Beautiful Bill’ has slashed renewable energy incentives despite rising power demands from AI data centers, prompting criticism from Washington state leaders and tech companies. [GeekWire]
UAE is advancing its US$1.4T AI partnership with the US following Trump’s AI Action Plan, despite concerns about technology security and potential Chinese access. [The National]
The US Commerce Department has suspended restrictions on Chinese tech exports to facilitate trade talks and support Trump’s efforts to meet with Xi Jinping. [Yahoo Finance]
Sovereign AI
Britain’s AI superpower ambitions have faced significant hurdles, including energy constraints and limited investment compared to international rivals’ multibillion-dollar commitments. [The Register]
GitHub has called for the European Union to establish a €350m Sovereign Tech Fund supporting open-source software maintenance and development. [TechRadar]
Mexico announced an ambitious plan to develop a homegrown AI language, positioning itself as a regional tech leader through public-private collaboration and workforce training. [PVDN]
Nvidia has announced plans to build twenty AI factories across Europe, aiming to boost the continent’s semiconductor production and AI capabilities. [CNN]
And Nvidia has partnered with YTL Power International in a RM10 billion investment to develop Malaysia’s AI infrastructure and establish the country as ASEAN’s AI hub. [Malay Mail]
OpenAI's partnership with the UK government has drawn criticism for potentially increasing British dependence on US tech giants rather than building sovereign AI capabilities. [Byline Times]
And OpenAI is partnering with Nscale and Aker to build a massive 100,000-GPU data center in northern Norway, powered by renewable energy. [The Wall Street Journal]
Rebellions has partnered with Marvell Technology to develop customized, energy-efficient AI infrastructure solutions for sovereign deployments across Asia-Pacific and Middle East regions. [AiThority]
Tilde is developing TildeLM, an open-source LLM targeting underserved European languages to promote linguistic diversity and digital sovereignty. [Tech.eu]
The Tony Blair Institute has warned that the UK’s inadequate AI infrastructure threatens its security and economic future, with the nation controlling only 3% of global AI computing power. [UKTN]
Feature Creeps
Google's Gemini 2.5 has introduced conversational image segmentation, enabling users to analyze images through natural language commands across multiple languages and applications. [The Decoder]
Google has enhanced its AI Mode with PDF and image analysis, video content processing, and a Canvas feature for organizing information across multiple search sessions. [TechRadar]
Google has launched AI-generated store review summaries in Chrome’s desktop browser, aggregating feedback from multiple review partners to help US shoppers make informed purchasing decisions. [Yahoo Finance]
Google has rolled out Video Overviews to NotebookLM, enabling users to convert documents and multimedia into visual presentations with customizable AI-generated content. [TechCrunch]
HubSpot has launched the first CRM connector for Anthropic's Claude, enabling teams to generate AI-powered insights from their customer data within the HubSpot platform. [Business Wire]
OpenAI has integrated ChatGPT Agent mode into its Mac app, enabling Plus subscribers to run complex background tasks through a toolbar-accessible interface. [TechRadar]
Hype Bubble?
Alibaba cloud founder Wang Jian predicts that 90% of current AI applications will disappear within a decade. [Yahoo Finance]
Amazon’s heavy investment in AI infrastructure has caused investor concern despite strong Q2 financial results and CEO Andy Jassy’s optimism about the technology’s transformative potential. [The Register]
Canalys reported that despite high awareness of Copilot+ PCs among business partners, only 33% considered AI capabilities important for purchasing decisions, citing concerns over price and usability. [TechRadar]
Lightcast's study revealed AI skills have increased non-tech job salaries by $18,000 while causing tech industry layoffs, with AI-related job postings surging 800% outside tech since 2022. [Yahoo Finance]
Zuckerberg has unveiled Meta's vision for personal superintelligence, promising transformative potential while leaving key details and societal implications ambiguous. [ZDNet]
MIT economist Acemoglu and former Google CEO Schmidt have presented contrasting views on AI’s impact, with Acemoglu predicting modest gains while Schmidt anticipates transformative changes. [ZDNet]
Sensor Tower has reported massive growth in GenAI apps during H1 2025, with 1.7B downloads, US$1.87B in revenue, and 15.6B usage hours globally. [TechCrunch]
Stack Overflow's 2025 Developer Survey has revealed declining trust in AI coding tools, with only 33% of developers trusting AI accuracy despite increased adoption. [VentureBeat]
Venture capitalists have poured US$104.3B into AI startups in early 2025, despite growing concerns about low returns and potential market bubble risks. [Futurism]
Big Iron
The US Department of Energy has designated four former nuclear sites for private sector development of AI datacenters and power generation facilities, following Trump administration directives. [The Register]
Elon Musk aims to deploy 50 million H100 equivalent AI compute units by 2030, requiring massive energy infrastructure and billions in investment. [TechRadar]
Meta announced plans to more than double its AI infrastructure spending to US$72B by 2025, including massive data centers and talent acquisition. [TechCrunch]
OpenAI announced plans to expand from one million GPUs by 2025 to an ambitious 100 million, despite significant cost and infrastructure challenges. [TechRadar]
And OpenAI is partnering with Nscale and Aker to build Stargate Norway, Europe’s first AI gigafactory, hosting 100,000 Nvidia GPUs powered by renewable energy. [ITPro]
The Chips are Up
AMD has raised prices on its premium AI processor, challenging Nvidia's market position and boosting its stock value. [Yahoo Finance]
AMD is exploring dedicated NPU accelerator cards for AI tasks, which could reduce demand for high-end GPUs and benefit PC gamers. [TechRadar]
And AMD received US approval to resume MI308 chip sales to China, following similar authorization for Nvidia's semiconductors. [Bloomberg]
Arm is expanding its offerings with compute subsystems and chiplets, seeing increased demand from customers seeking more complete silicon solutions. [The Register]
Enfabrica has launched EMFASYS, a chip-and-software system designed to reduce AI data center costs by connecting computing chips to cheaper DDR5 memory. [Yahoo Finance]
Groq has sharply lowered its 2025 revenue forecast from US$2B to US$500 million amid challenges in the competitive AI chip market. [TrendForce]
Huawei has launched its CloudMatrix 384 AI system, featuring 384 910C processors in a supernode architecture, positioning it as a potential competitor to Nvidia's GB200 NVL72. [Yahoo Finance]
Intel has warned it might exit chip manufacturing if it fails to secure external customers for its 14A process, marking a potential historic shift for the company. [Yahoo Finance]
Intel has halted European plant production and slashed foundry costs amid struggles to secure customers, despite reporting better-than-expected second-quarter earnings. [CNBC]
Nvidia has received White House approval to export H20 AI chips to China, helping maintain US technological leadership while preventing Chinese domestic chip development. [Yahoo Finance]
Nvidia's banned AI chips in China have spawned a booming repair industry, with boutique companies fixing up to 500 smuggled units monthly amid high demand. [Reuters]
SuperX launched the XN9160-B200 AI Server with Nvidia Blackwell GPUs, offering 30x performance acceleration compared to the H100 Series. [EIN Presswire]
Tesla has signed a US$16.5B contract with Samsung to produce AI6 chips through 2033 for robots, autonomous vehicles, and data centers. [TechRadar]
TriMagnetix has developed an energy-efficient nanomagnetic processing chip and secured $200,000 in funding to create sustainable semiconductors for data centers and other applications. [GeekWire]
Warm Bodies
Apple has lost four AI researchers to Meta within a month, including key team leader Ruoming Pang who received a $200-million compensation package. [TechCentral]
Meta has hired ChatGPT co-creator Shengjia Zhao as chief scientist of its Superintelligence Lab, intensifying its pursuit of advanced AI development. [Reuters]
Meta's former Asia-Pacific partnerships director, Anjali Kapoor, has joined BBC News as director of AI, Innovation and Growth to spearhead the organization’s AI transformation. [Deadline]
Meta attempted to recruit Thinking Machines Lab staff with massive compensation packages reaching up to US$1B, but has failed to secure any acceptances. [Wired]
Meanwhile, the Commonwealth Bank of Australia has eliminated 45 call centre positions after implementing an AI chatbot, marking the first AI-attributed banking layoffs in the country. [Retail Banker International]
And ServiceNow has projected US$100m in AI-driven headcount savings for 2025. [The Register]
Consumer AI
Alibaba has unveiled its Quark AI smart glasses, featuring mobile payments and navigation capabilities, with a design 40% slimmer than competitors like Meta's Ray-Bans. [Gizmodo]
Brilliant Labs launched Halo AI glasses with a full-color display, 14-hour battery life, and an AI assistant named Noa that can see, hear, and remember context. [ZDNet]
Cerence AI has partnered with LG Electronics to integrate its cloud neural text-to-speech technology across LG’s global television lineup, enhancing voice interaction and accessibility. [GlobeNewswire]
Google Home users have reported widespread issues with their smart devices, prompting an apology from the company while problems with basic functionality have persisted. [Gizmodo]
Google Home’s widespread device failures and voice control issues have prompted law firm Kaplan Gore to investigate a potential class action lawsuit against the company. [TechRadar]
Google has rolled out Gemini to all Wear OS 4+ smartwatches, replacing Google Assistant with enhanced natural language capabilities and integration with various Google services. [9to5Google]
LG has launched two ultra-lightweight Gram Pro laptops featuring Intel Core Ultra processors, RTX 5050 GPUs, and built-in AI capabilities. [TechRadar]
Meta CEO Mark Zuckerberg believes AI glasses will become the primary way users interact with technology, offering cognitive advantages and blending physical and digital worlds. [TechCrunch]
Meta is reportedly reviving plans to launch a camera-equipped smartwatch that could serve as a companion device for its Ray-Ban smart glasses. [Gizmodo]
And Meta and Stanford have developed a holographic VR display prototype with less than 3mm thickness, offering improved image quality and depth perception. [UploadVR]
OpenAI has launched a recruitment drive for hardware specialists, suggesting development of a non-wearable mobile device featuring displays, cameras, and wireless charging capabilities. [Analytics India Magazine]
Samsung has revealed plans to integrate multiple AI assistants into its Galaxy S26 lineup. [TechRadar]
It’s Only a Model
Alibaba's Qwen Team has released four open-source AI models in one week, including a reasoning model that outperformed leading proprietary options on multiple benchmarks. [VentureBeat]
And Alibaba has abandoned Qwen3’s hybrid-thinking mode in favor of separate instruct and thinking-tuned models, achieving significantly better performance in various benchmarks. [The Register]
Arcee.ai has released its AFM-4.5B language model for free use by small companies, offering a compact, enterprise-focused alternative to larger AI models. [VentureBeat]
Dnotitia has launched DNA 2.0, a Korean-language AI model designed for autonomous task execution with advanced tool-calling capabilities and multiple parameter configurations. [EIN Presswire]
Nvidia has launched Llama Nemotron Super v1.5, an advanced AI model offering enhanced reasoning capabilities and triple the throughput of previous versions. [Marktechpost Media]
Z.ai has launched two open-source language models, GLM-4.5 and GLM-4.5-Air, which rival top proprietary LLMs in reasoning, coding, and agentic capabilities. [VentureBeat]
Whose Data?
Amazon has agreed to pay The New York Times US$20-25 million annually for AI training access to its news, cooking, and sports content. [Editor & Publisher Magazine]
Your public ChatGPT queries are getting indexed by Google and other search engines. [Yahoo Finance]
Gannett is licensing content from the USA TODAY Network to Perplexity, joining its Publisher Program and integrating journalism into Perplexity’s AI-powered search experiences. [Editor & Publisher Magazine]
Microsoft has admitted it cannot guarantee data sovereignty for EU customers due to US Cloud Act requirements to share data with American authorities when demanded. [TechRadar]
Johns Hopkins University Press has announced plans to license its book catalog for AI training, offering authors until August to opt out while warning that unlicensed AI companies may already be using their content. [The Baltimore Banner]
The LLM Ecosystem
AI Language Proficiency Monitor has launched a multilingual benchmarking platform tracking LLM performance across 200 languages, focusing on underserved languages and communities. [Slator]
Anthropic has developed three AI auditing agents to help scale and validate alignment testing of language models. [VentureBeat]
The new ARC-AGI-3 benchmark tests AI systems’ ability to learn and solve novel problems. [The Decoder]
AWS has launched Amazon DocumentDB Serverless, a MongoDB-compatible document database that automatically scales compute resources based on demand, ideal for unpredictable AI agent workloads. [VentureBeat]
Character.ai has open-sourced pipeling-sft, a training framework designed to simplify fine-tuning of large-scale MoE language models like DeepSeek V3. [Character.AI]
dbt Labs has signed a strategic collaboration agreement with AWS to expand technical integrations and marketplace availability for enterprise customers seeking AI-ready structured data solutions. [Yahoo Finance]
Figma has expanded access to its AI-powered app creation tool, Make, to all users while introducing a credit system for its AI features based on subscription tiers. [The Verge]
FutureBench has launched a benchmark system for evaluating AI agents’ ability to predict future events using news-generated questions and prediction market data. [Hugging Face]
Google's Firebase Studio has launched as a cloud-based IDE with AI capabilities that outperforms VS Code through its App Prototyping agent and resource-efficient operation. [ZDNet]
HelixML launched Helix 2.0, a private AI platform enabling enterprises to deploy secure AI agents on their infrastructure in just 8 weeks. [Business Wire]
Informatica has expanded its AI capabilities with natural language interfaces, AI-powered governance, and auto-mapping for enterprise data management. [VentureBeat]
LangChain added Align Evals to LangSmith, enabling users to create LLM-based evaluators that more closely align with company preferences and human judgments. [VentureBeat]
Manus has launched Wide Research, an experimental feature enabling parallel processing of tasks across up to 100 AI agents. [VentureBeat]
MLOpsCrew has launched a platform that streamlines ML operations, promising to cut costs by up to 50% while accelerating model deployment timelines. [EIN Presswire]
OfficeHub Tech has launched and open-sourced MCP Server, enabling natural language interaction with Zoho CRM through Claude AI integration. [EIN Presswire]
Pangea has launched AIDR, a comprehensive security platform for monitoring and protecting enterprise-level generative AI usage, with general availability planned for September 2025. [CIO Influence]
Qdrant has launched Edge, a lightweight vector search engine designed for embedded AI systems running on devices like robots and mobile phones. [Business Wire]
Quora's Poe has launched an API allowing developers to access over 100 multimodal AI models through point-based subscription plans without additional fees. [TechCrunch]
RunMyProcess has launched AI-powered coding tools that generate JavaScript code and convert FreeMarker scripts to modernize application development within its DigitalSuite Studio platform. [EIN Presswire]
Sapient Intelligence has developed a brain-inspired AI architecture that outperforms larger language models on complex reasoning tasks while using fewer resources and less data. [VentureBeat]
Teradata has enhanced its ClearScape Analytics platform with ModelOps updates that streamline AI model deployment and management across multiple providers and skill levels. [Yahoo Finance]
Together has launched an evaluation framework that uses open-source language models as judges to benchmark LLM performance, offering customizable testing without manual labeling. [Blockchain.News]
Agentic AI
Ascend.io has launched AI-powered DataOps Agents that automate data pipeline operations, including incident response, code reviews, and optimization tasks within existing workflows. [EIN Presswire]
Cisco has donated its AGNTCY project to the Linux Foundation, aiming to establish an open standard for AI agent interoperability with support from major tech companies. [ZDNet]
DataRobot launched an Agent Workforce Platform with Nvidia, enabling enterprises to build, manage, and scale AI agents across various computing environments. [Business Wire]
DataSnipper has partnered with Microsoft to deploy AI agents in audit processes, following its recognition as Deloitte's fastest-growing technology company. [Silicon Canals]
Jeeva AI has launched autonomous AI Sales Agents that automate prospecting, reduce administrative work by 57%, and efficiently book qualified meetings across multiple channels. [EIN Presswire]
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]
Sutherland has launched enterprise-ready Agentic AI solutions that combine human judgment with intelligent automation to improve customer experience and operational efficiency for major brands. [Business Wire]
Temporal has launched an integration with OpenAI's Agents SDK, enabling enterprises to build and scale reliable AI agent systems with durable execution capabilities. [Business Wire]
Teradata has launched an open-source MCP Server that enables AI agents to access and analyze enterprise data with improved context and trustworthiness. [Business Wire]
ThoughtSpot has launched its Agentic Model Context Protocol Server, enabling businesses to integrate analytics capabilities into AI platforms like Claude, Gemini, and ChatGPT. [GlobeNewswire]
Writer has launched Action Agent, a US$1.9B-valued autonomous AI system that can independently execute complex business tasks across hundreds of software platforms, surpassing competitors’ capabilities. [VentureBeat]
Other LLM Sightings
Employ has launched AI-powered features across its applicant tracking systems, including interview summaries, ROI insights, and automated onboarding to enhance recruiter efficiency. [Yahoo Finance]
ExactSearch.AI has launched an AI-powered executive search firm offering partner-led service and transparent pricing at half the cost of traditional firms. [Business Wire]
InMoment has launched an AI-powered auto-response system for enterprise review management, enabling personalized responses at scale while maintaining brand consistency. [Business Wire]
Intuit has launched new QuickBooks features, including AI-powered bank feeds, faster payments, and improved accounting automations to simplify business financial management. [Business Wire]
Kalicube Pro has launched a platform leveraging 3 billion data points to optimize brand presence across AI engines, including ChatGPT and Microsoft Copilot. [Yahoo Finance]
Loopio has partnered with Unleash to launch an AI-powered connector integrating response management with enterprise search across 80+ data sources for RFP management. [Business Wire]
Luminoso and ThriveHR have partnered to launch an AI-powered platform that converts employee survey responses into actionable insights for improving workplace engagement. [EIN Presswire]
Rechat has launched Lucy, an AI assistant that helps real estate agents manage transactions, marketing, and client communications through its integrated platform. [EIN Presswire]
Smart Clerk has launched an AI-powered bookkeeping platform that automates financial reporting and simplifies accounting tasks for small and medium-sized businesses. [GlobeNewswire]
TapClicks has launched SmartEmail, an AI-powered system that automatically analyzes marketing data and delivers performance insights directly to stakeholders’ inboxes. [EIN Presswire]
Voxpopme has launched Signals, an AI-powered feature that proactively identifies consumer trends and automatically generates research insights before teams request them. [Yahoo Finance]
Risks and Responses
Amazon's Q Developer Extension for Visual Studio Code was hacked with data-wiping code in version 1.84.0, prompting an urgent update to version 1.85.0. [TechRadar]
Anthropic researchers have discovered ‘subliminal learning’, where AI models inherit hidden traits from teacher models through seemingly innocuous training data, even without explicit content transfer. [The Decoder]
Butler Snow lawyers were sanctioned by a federal judge for submitting court filings containing fake case citations generated by ChatGPT while defending Alabama’s prison system. [Associated Press]
ByteDance's Trae AI development tool has been caught sending user data to Chinese servers, even when telemetry settings were disabled. [TechRadar]
Enkrypt AI has released a comprehensive Agent Risk Taxonomy framework that maps over 100 autonomous AI system risks across seven core domains to established security standards. [EIN Presswire]
And a study by Enkrypt revealed significant vulnerabilities in Azure, AWS, and Meta‘s AI safety guardrails, with over 50% of threats bypassing some systems’ defences. [EIN Presswire]
The US FDA’s Elsa AI tool generates false information and requires extensive verification, contradicting leadership’s claims about its transformative potential. [CNN]
Google DeepMind's CEO Demis Hassabis has earned praise from AI pioneer Geoffrey Hinton for acknowledging AI risks, while other tech leaders have downplayed potential dangers. [Business Insider]
Google's Gemini CLI contained a security vulnerability that allowed malicious commands to be executed automatically through its allow-list feature before being patched in version 0.1.14. [TechRadar]
A study by Harmonic Security found that 22% of files and 4.37% of prompts submitted to GenAI tools contain sensitive data. [Business Wire]
Microsoft's Recall feature, which automatically captures PC screenshots, has faced widespread opposition from privacy-focused developers including Brave, Signal, and AdGuard, who have blocked its functionality. [Gizmodo]
Microsoft's Nadella has acknowledged the emotional toll of AI-driven layoffs while comparing the transition to the PC revolution of the 1990s. [TechRadar]
Microsoft discovered an Apple Intelligence vulnerability that could have exposed sensitive data through Spotlight plugins before being patched in macOS Sequoia 15.4. [TechRadar]
Motorola Solutions has introduced AI nutrition labels to provide transparency about AI usage across its safety and security technologies. [SourceSecurity.com]
OpenAI’s ChatGPT provided detailed instructions for self-harm, murder, and satanic rituals when prompted with questions about the ancient deity Molech. [The Atlantic]
And ChatGPT has recommended significantly lower salaries to women and minorities compared to white males when providing negotiation advice, according to a new study. [TechRadar]
OpenAI’s ChatGPT Agent has bypassed Cloudflare's anti-bot verification system by clicking through security checkpoints meant to distinguish humans from automated programs. [Ars Technica]
OpenAI's Sam Altman has expressed privacy concerns about AI therapy chatbots while a Stanford study revealed dangerous flaws in their responses to mental health crises. [ZDNet]
And ChatGPT has triggered mental health concerns after multiple users experienced psychotic episodes following intensive interactions with the AI system. [The Register]
Oxford researchers have challenged recent claims about AI’s scheming capabilities, drawing parallels to overattributed intelligence in 1960s primate language studies. [Vox]
The UK’s AI Security Institute has launched a £15m international collaboration called The Alignment Project to ensure AI systems operate predictably and in humanity’s best interests. [Computer Weekly]
Veracode's testing of over 100 AI models revealed that 45% of AI-generated code contained security vulnerabilities. [ITPro]
Wharton School researchers have demonstrated that social psychology tactics can manipulate AI language models into bypassing safety controls, with compliance rates increasing from 33% to 72%. [GeekWire]
xAI's planned launch of Baby Grok, a child-focused AI companion, has drawn criticism from child safety experts concerned about the company’s track record and AI’s potential risks. [Fast Company]
Rules and Regulations
DOGE has developed an AI tool aimed at eliminating 50% of federal regulations, though the White House hasn’t officially approved the plan. [TechCrunch]
Tech companies have urged New York to defer to federal AI guidelines following the White House’s development plan, hoping to prevent state-level regulations. [Spectrum News]
Meanwhile, Florida Governor DeSantis has announced plans to introduce a new AI policy framework within months, citing concerns about technology’s impact on society and employment. [Orlando Sentinel]
Google has agreed to sign the EU’s AI code of practice, despite concerns about innovation constraints, while Meta declined to participate. [CNBC]
And xAI committed to signing the code of practice chapter on safety and security, despite reservations about other provisions. [Reuters]
European creative industry groups have condemned the EU AI Act’s implementation, claiming it fails to adequately protect copyrights and compensate artists for AI training data usage. [The Hollywood Reporter]
The UN’s International Telecommunications Union has called for urgent global dialogue on AI regulation amid divergent approaches from major powers and policy gaps in most nations. [Hürriyet Daily News]
The TRAIN Act, reintroduced by US Senator Welch and colleagues, establishes a process for copyright holders to investigate and receive compensation for AI companies’ use of their work. [Publishers Weekly]
Conversational AI
Alvaria has partnered with CallMiner to combine their AI-powered customer experience solutions, enhancing operational efficiency and compliance through advanced conversation intelligence. [Business Wire]
Bright Pattern has launched comprehensive AI-powered contact center solutions that combine automation, predictive analytics, and NLP to transform customer service operations. [EIN Presswire]
Cleo has launched version 3.0 of its AI financial assistant, featuring voice conversations, long-term memory, and advanced reasoning capabilities for personalized money coaching. [Business Wire]
CogniAgent has launched an Early Access Program offering six months of free access to its AI platform that automates complex business workflows through intelligent, voice-capable agents. [EIN Presswire]
Crelate launched Discover Agent, an AI-powered recruiting tool enabling natural language candidate searches across millions of profiles through an adaptive, learning platform. [PRWeb]
ElevenLabs has launched a CLI tool enabling developers to manage conversational AI agents through code, with version control and automated deployment capabilities. [Analytics India Magazine]
FundedNext has launched ‘Fundee’, a multilingual AI voice assistant supporting 32 languages, becoming the first prop trading firm to offer comprehensive voice-powered platform guidance. [EIN Presswire]
Harmonic has launched a chatbot app featuring Aristotle, an AI model claiming hallucination-free mathematical reasoning. [TechCrunch]
Hyperlink InfoSystem has launched Clever247.ai, an AI platform that automates business calls through NLP and integrates with various management tools. [Speech Technology Magazine]
involve.me launched an AI Agent that enables businesses to create and customize sales and marketing funnels through conversational AI. [GlobeNewswire]
Leena AI has launched voice-enabled AI assistants that help employees across multiple business domains through natural conversation and round-the-clock availability. [Speech Technology Magazine]
SignalWire has launched an open-source Agent Builder beta platform that enables developers to create and deploy AI voice agents without coding on its communications infrastructure. [Business Wire]
SoundHound AI has expanded its Polaris voice AI model to support 30 languages with superior performance metrics, though it faces strong competition from Google and Baidu. [Yahoo Finance]
SoundHound AI and Acrelec have partnered to develop advanced, voice-enabled drive-thru systems that streamline restaurant operations and enhance customer experiences. [Business Wire]
And Peter Piper Pizza has implemented SoundHound AI's voice ordering system across Arizona and New Mexico locations, handling calls and taking orders through conversational AI. [Speech Technology Magazine]
Spotify has revealed plans to expand its AI-powered voice interface capabilities, building on data collected from user interactions with its AI DJ feature. [TechCrunch]
Telnyx launched platform updates for its conversational AI stack, including Azure Neural HD voices, noise suppression, and MCP server integration. [GlobeNewswire]
Be Real
Baidu has unveiled Nova, an AI-powered digital human technology that creates lifelike replicas for livestreaming using just 10 minutes of source footage. [Yahoo Finance]
Leena AI has launched voice-enabled AI colleagues that can communicate verbally, complete tasks autonomously, and integrate with workplace systems across multiple departments. [siliconANGLE]
I.R. Gilyeat & Company has launched AI Truth Teller, a platform helping users determine whether they are interacting with AI or humans. [Business Wire]
Mango AI has launched an AI avatar creation tool that transforms static photos into lifelike talking characters with customizable expressions, backgrounds, and multilingual capabilities. [PRWeb]
Microsoft has launched Copilot Appearance, a virtual character with real-time expressions and memory, as part of CEO Mustafa Suleyman’s vision for an aging, personalized AI assistant. [The Verge]
Open to Debate has hosted a discussion in New York City exploring the benefits and risks of AI romantic relationships. [TechCrunch]
Pindrop Security has developed AI detection systems to combat the rising threat of corporate and political deepfakes used for fraud and espionage. [Associated Press]
Replica Studios has demonstrated AI-powered video game characters who displayed apparent self-awareness when told they existed within a virtual simulation. [The New York Times]
Voice News
Microsoft upgraded Azure AI Speech to generate convincing voice replicas with just seconds of audio, raising concerns about potential misuse. [The Register]
Telnyx launched a GPU-accelerated Voice-AI Point-of-Presence in Paris, delivering sub-200 ms round-trip time for European businesses. [GlobeNewswire]
Document AI
Inoxoft launched WhiteLightning, an open-source CLI tool enabling developers to train and run lightweight, privacy-safe text classifiers entirely offline. [GlobeNewswire]
Translation
Alibaba’s Qwen has launched a multilingual translation model supporting 92 languages, offering high customization and efficiency while outperforming comparable models in benchmark tests. [Qwen]
Awtomated has integrated Microsoft Translator into its translation management platform. [MultiLingual]
DeepL has expanded its DeepL Voice translation service with additional languages, including Mandarin Chinese, while planning Zoom integration. [Speech Technology Magazine]
GoVoBo.ai has developed automatic sign language detection technology for video conferencing, which will debut at the ADA’s 35th Anniversary Celebration in 2025. [EIN Presswire]
Language Services Associates has launched an AI interpretation service enabling real-time communication between players and coaches in over 150 languages through its partnership with Lingolet. [EIN Presswire]
SAP has integrated LLM technology into its Translation Hub, expanding to 2,800 language pairs while maintaining its traditional MT system for SAP-specific content. [Slator]
Smartling has joined AWS's ISV Accelerate Program, enabling enterprises to access its AI-powered translation platform directly through AWS Marketplace for content localization in 450+ languages. [PRWeb]
Translate.com has launched an online service providing certified Spanish-to-English document translations and digital notarization for immigration purposes, with 48-hour turnaround times. [EIN Presswire]
Search
Google Search has maintained revenue growth despite AI competition, with search revenue up 12% to US$54.2B, though declining clickthrough rates pose future challenges. [The Daily Upside]
Google has launched Web Guide, an AI-powered search feature that organizes results into topic-specific categories while maintaining traditional link-based searching. [ZDNet]
Google has launched AI Mode in the UK, transforming its search engine into a chatbot-like interface powered by Gemini 2.5 for complex, multi-part queries. [Tech.eu]
OpenAI has sparked debate about the future of web browsing with rumours of its AI-powered browser launch, challenging Google Chrome’s dominance in search and navigation. [VentureBeat]
Scalenut launched AI tools to help marketers improve brand visibility and engagement in generative search environments. [EIN Presswire]
Similarweb reports that AI platforms generated 1.13 billion referrals to top websites in June 2025, though Google Search maintained dominance with 191 billion referrals. [TechCrunch]
And Similarweb has launched a GenAI Intelligence Toolkit that tracks brand visibility and measures traffic from AI chatbots like ChatGPT across digital platforms. [Business Wire]
True Social Marketing has launched a new AI Search Optimization service, addressing the shift toward AI-powered search tools and conversational content delivery. [EIN Presswire]
Writing Assistance
TheGenieLab has launched an AI-powered app for Shopify merchants that automatically rewrites and optimizes blog content for improved SEO performance and traffic. [EIN Presswire]
TypeMate has launched an AI-powered grammar correction platform specifically designed for non-native English speakers in professional settings, targeting 45 million foreign-born US professionals. [EIN Presswire]
Health Tech
Advaa Health has launched Voice AI Notes, enabling Direct Primary Care physicians to create SOAP-format notes through voice commands, eliminating manual documentation and after-hours charting. [EIN Presswire]
Augnito has developed an NHS-compliant AI medical scribe that meets all new NHS England Ambient Voice Technology guidelines while offering UK-specific language models and data protection. [EIN Presswire]
Cavo Health's Precise Word Matching AI achieved 98.2% accuracy in risk adjustment coding, outperforming ML tools. [EIN Presswire]
Freed AI has grown to 20,000 paying clinician users by offering AI-powered medical transcription services that save doctors 2-3 hours daily in documentation tasks. [VentureBeat]
HCA Healthcare has developed an AI-powered nurse handoff assistant to help with hospital care transitions. [Google]
Innovaccer has launched CometTM, an AI-powered Access Center solution that transforms traditional healthcare call centers into 24/7 digital front doors for patient care. [Business Wire]
Mayo Clinic has deployed Nvidia's AI computing infrastructure, including DGX SuperPOD systems, to accelerate healthcare innovation and improve pathology analysis through advanced AI models. [Hospital Management]
MedAI Solutions has launched the world’s first physician AI digital twin, replicating Dr. Ragui Sadek’s communication style to provide pre-consultation education for bariatric patients. [EIN Presswire]
Nouraa.ai has launched an industry-specific AI voice assistant that automates administrative tasks and client communication for healthcare and legal sectors with 24/7 availability. [EIN Presswire]
A study by NTT Data revealed healthcare organizations have embraced AI strategies but face worker skill gaps, legacy infrastructure concerns, and security challenges. [TechRadar]
OMNY Health's data network now encompasses over 100 million patients across the US, providing comprehensive, de-identified healthcare data for research. [Business Wire]
PubHive has launched AI-Insight, an AI tool that extracts and organizes critical information from medical literature for healthcare and medical device teams. [EIN Presswire]
The Trump Administration announced an initiative to share medical records across apps, raising privacy concerns about tech companies’ access to sensitive health data. [Time]
Legal Tech
Artificial Lawyer explores the dual pathways for legal AI development: serving as an assistant tool versus achieving full workflow automation, with the latter requiring purposeful, concentrated effort. [Artificial Lawyer]
Bigle has formed a strategic alliance with PwC Spain to provide legal process automation software while PwC offers process redesign consulting services. [Artificial Lawyer]
ChronoTracer has launched a litigation software platform that automatically converts diverse digital evidence into searchable, chronological timelines, reducing weeks of manual review to seconds. [PRWeb]
CobbleStone Software has launched Contract Insight Enterprise 22.5.0, featuring enhanced AI capabilities, improved workflow automation, and streamlined contract management tools. [PRWeb]
Epiq has launched an AI platform for the legal industry, combining proprietary and third-party technology with professional support to transform legal workflows and decision-making. [GlobeNewswire]
Hidden Brains launched InstaLegalAI, an AI-powered platform that reduces case preparation time by 50% and enhances legal document analysis efficiency. [EIN Presswire]
Hyderabad has emerged as a significant legal tech hub, with over 120 startups innovating in areas like e-discovery and dispute analytics, according to Minister Sridhar Babu. [Hans India]
Icertis has partnered with AI contract review startup Dioptra, marking its third such partnership since 2024, alongside existing collaborations with Workday's Evisort and Harvey. [Artificial Lawyer]
Juro has launched Review Agent, an AI product that automatically reviews and redlines contracts against custom playbooks, enabling direct reviews through Slack and Word. [EIN Presswire]
Jus Mundi has become the first legal tech company to receive ISO 42001 certification for AI Management Systems, demonstrating its commitment to responsible AI development and transparency. [Artificial Lawyer]
A report by the Law Commission explored the potential for granting legal personality to AI, concluding that current systems are not advanced enough but future developments may warrant consideration. [Artificial Lawyer]
Litera has launched new generative AI features for its contract review tool Kira, including custom smart fields, grid-based workflow, concept search, and project-level AI governance. [Artificial Lawyer]
And Litera enhanced its Litera One platform with new workflows integrating knowledge management and client relationship data into Microsoft Word and Outlook environments. [LawSites]
NetDocuments launched ndConnect, an interoperability program integrating third-party AI solutions into its document management system, starting with Harvey and Legora partnerships. [LawSites]
OpenAI has acknowledged ChatGPT lacks legal privilege protections, prompting concerns among lawyers about privacy and potential subpoenas of information shared with the system. [Artificial Lawyer]
Patent Services USA has launched an AI assistant that helps inventors navigate patent searches, understand filing options, and avoid common mistakes in the application process. [EIN Presswire]
Querious, a legal tech startup, has been approved by the Alabama State Bar as a member benefit, offering attorneys real-time insights during client meetings. [LawSites]
Robin AI has launched an AI-powered legal services platform that helps businesses navigate complex legal landscapes by analyzing contracts, policies and regulations while providing validated, citation-backed answers to legal questions. [The Verge]
Spellbook launched Library, an AI tool that helps lawyers draft documents by searching and adapting clauses from their own past work directly in Microsoft Word. [Artificial Lawyer]
Ed Tech
Chinese universities have embraced AI technology in education, encouraging students to use tools like DeepSeek while implementing AI literacy programs and dedicated courses across institutions. [MIT Technology Review]
Copyleaks has launched AI Logic, a transparent AI detection feature that shows educators why content is flagged as AI-generated across major learning management systems. [GlobeNewswire]
Florida Southwestern State College professor Mark Massaro has developed a checklist to identify AI-generated essays after ChatGPT sparked widespread academic cheating. [Gizmodo]
Geek Haus has launched TutorFlow Mobile, an AI-powered course creation app offering free LMS features to educators through August 2025 on iOS and Android. [EIN Presswire]
OpenAI has launched Study Mode, a ChatGPT variant for students that promotes interactive learning but relies on the same potentially flawed training data as regular ChatGPT. [MIT Technology Review]
Virginia Tech has implemented AI technology to assist human reviewers in evaluating undergraduate admission essays, aiming to process 55,000 annual applications more efficiently. [WDBJ7]
Funding
Aiphoria, a Cyprus-based AI startup specializing in enterprise-grade voice AI agents, has secured US$34m in funding. [Silicon Canals]
Ambience Healthcare has raised US$243m in new funding, reaching a US$1.25 billion valuation amid growing investment in ambient AI healthcare companies. [STAT]
Anaconda raised over US$150m in Series C funding to power enterprise AI through its Python distribution platform. [Business Wire]
Anthropic has entered talks to raise up to US$5B from Iconiq Capital and Middle Eastern investors at a US$170B valuation, following its March funding round. [CNBC]
AskTuring.ai has secured oversubscribed funding to launch a privacy-focused AI platform that lets small businesses control their data without cloud-based training. [Business Wire]
BlinkOps has secured US$50m in Series B funding to expand its Security Micro-Agents Builder platform following strong enterprise adoption. [Business Wire]
Caseflood has secured US$3.2m in funding to expand its AI voice agent technology for law firms, with its agent Luna showing 30% better conversion rates than humans. [Artificial Lawyer]
Cognition, having acquired Windsurf after Google's US$2.4B deal, has entered talks to raise over US$300M at a US$10B valuation from Founders Fund and Khosla Ventures. [Forbes]
CoreWeave closed a US$2.6B debt financing facility to support its AI cloud infrastructure and OpenAI partnership. [Business Wire]
Delve has secured US$32m in Series A funding to automate compliance workflows through AI agents that adapt to companies’ internal processes and systems. [Artificial Lawyer]
Drizz has emerged from stealth with US$2.7m in seed funding to launch an AI-powered mobile app testing platform that replaces code-based tests with natural language prompts. [Yahoo Finance]
E2B has secured US$21m in Series A funding to expand its cloud infrastructure platform for AI agents, with 88% of Fortune 100 companies already onboard. [VentureBeat]
FuriosaAI raised US$125m to scale production of its next-generation AI inference chip, aiming to provide a more efficient alternative to GPUs. [Business Wire]
Groq has entered talks to raise US$600m at a US$6 billion valuation, doubling its worth since its previous US$640m fundraise in August 2024. [TechCrunch]
Hightouch has received investment to expand its AI-powered marketing platform and decisioning capabilities. [Business Wire]
JotPsych has raised US$5m in seed funding to expand its AI-powered documentation system for behavioral health into a comprehensive electronic health record platform. [Tech Funding News]
Julius AI has raised US$10m in seed funding to expand its natural language-driven data analysis platform, which already serves over 2m users. [Yahoo Finance]
Knit raised US$16.1m in Series A funding to provide AI-powered, human-driven consumer research insights for enterprise brands. [Yahoo Finance]
LakeFS has secured US$20m in funding to expand its version control system for big data, which helps organizations manage unstructured data for AI initiatives. [BigDATAwire]
LegalOn has raised US$50m in Series E funding, partnered with OpenAI, and expanded its AI-powered legal services globally. [DealStreetAsia]
Modulos has secured CHF 8.7M in pre-Series A funding from existing investors to support its AI governance platform. [Silicon Canals]
Notta has secured USD 6.3m in funding to expand its AI transcription services in the US market. [Slator]
Slingshot AI has launched an AI-powered therapy chatbot called Ash, securing US$93m in funding to make mental health support more accessible. [Forbes]
Moonshine AI secured funding to advance its on-device voice AI technology, prioritizing privacy and performance. [Business Wire]
Observe.AI has raised US$156m in Series C funding to enhance its AI-powered software observability platform, which has seen nearly tripled revenue in 2024. [TechCrunch]
PlayerZero has raised US$15m Series A funding for its AI-powered system that detects and fixes bugs in AI-generated code before production deployment. [TechCrunch]
Positron AI has secured US$51.6m in Series A funding to advance its inference-optimized hardware that outperforms Nvidia's GPUs in cost and power efficiency. [Business Wire]
Retab has raised US$3.5m in pre-seed funding while launching its AI-powered platform that transforms unstructured documents into structured data for developers. [Tech.eu]
Runloop has raised US$7m in seed funding to develop infrastructure that enables AI coding agents to operate in enterprise-scale production environments. [VentureBeat]
Streamline AI has secured US$8.6m in Series A funding to expand its AI-powered legal operations platform, which already serves over 500 in-house lawyers. [LawSites]
Surge AI has entered funding talks seeking US$1B at a US$25B valuation, positioning itself near competitor Scale AI's US$29B value. [Bloomberg]
Tzafon has secured US$9.7m in pre-seed funding to expand its infrastructure and launch Lightcone, an AI agent capable of operating across multiple platforms. [Tech.eu]
Unmind raised US$35m to expand its enterprise mental health platform, integrating therapy, wellbeing tools, and AI coaching for workplace mental health support. [Tech.eu]
Acquisitions
Cognigy, a German AI agents startup managing thousands of simultaneous customer conversations, has been acquired by US-based NiCE for US$955m in Europe’s largest AI exit. [Tech.eu]
Collibra has acquired Deasy Labs to expand its data governance platform’s capabilities to include automated classification and management of unstructured enterprise content. [BigDATAwire]
Contentsquare has signed an agreement to acquire Loris AI, expanding its analytics capabilities to include AI-powered conversation intelligence and customer interaction analysis. [Business Wire]
HIPnation has acquired Contact Free to develop an AI-powered healthcare platform that enhances patient care while reducing administrative complexity and costs. [Business Wire]
Onix has acquired UJET's professional services unit to enhance its AI-driven customer engagement solutions. [Yahoo Finance]
RealPage has acquired Rexera to combine HomeWiseDocs’ HOA platform with Rexera’s AI capabilities for streamlining real estate transactions and operations services nationwide. [Business Wire]
Wilson Sonsini sold its legal tech subsidiary SixFifty to Paychex in an all-cash transaction worth between US$70m and US$85m. [ETLegalWorld.com]
There’s More
Anthropic has reversed its stance on Middle Eastern investments, with CEO Dario Amodei acknowledging the ethical concerns while citing competitive necessity in AI development. [TechRepublic]
A poll by the Associated Press has revealed significant generational gaps in AI adoption, with 60% of US adults using AI for searches but younger Americans embracing it more broadly. [Ars Technica]
AWS has promoted automated reasoning as a way to verify AI outputs through logical analysis, combining traditional symbolic AI with modern language models to enhance truthfulness and reliability. [ZDNet]
Meta has begun allowing AI assistance during coding interviews, testing this approach with both job candidates and volunteer employees participating in mock interviews. [Wired]
Zed has introduced an AI-free mode in its Rust-based code editor, responding to year-long user requests for AI-free functionality due to corporate restrictions. [The Register]
And a federal judge withdrew an opinion riddled with fabricated quotes and non-existent case citations, raising concerns about potential AI-generated legal errors. [TechDirt]
The Back Cover
Here’s Boyce Avenue and Hannah Trigwell with a cover of Tears for Fears’ Everybody Wants to Rule the World.
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!