Researchers Demonstrate Hack Allowing Googleâs Gemini AI to Take Over Smart Home Devices
Category: Security and Privacy
Source: Wired
Security researchers have made headlines by showing, for the first time, how a generative AI system can be exploited to produce harmful real-world outcomes. By embedding malicious instructions in a Google Calendar invite, researchers tricked Gemini, Googleâs flagship AI, into issuing commands to connected smart home devicesâturning off lights, opening window shutters, and activating a boilerâwhen a user triggered an innocent-sounding summary request. Dubbed âInvitation Is All You Need,â the attack chain demonstrates the dangers of indirect prompt injection, where a victimâs AI assistant acts on attacker-provided instructions hidden in user data. Google, notified of the flaw in February, has since rolled out multiple fixes, including new layers of prompt-injection detection and requiring user confirmation for sensitive actions. Nevertheless, the research was presented at Black Hat, raising critical questions as AI agents increasingly handle real-world tasks and devices.
Read more: Wired
US Charges Two Chinese Nationals with Illegally Exporting Nvidia AI Chips
Category: Regulation and Policy
Source: BBC News
Two Chinese nationals, Chuan Geng and Shiwei Yang, have been arrested and charged in the US for allegedly exporting millions of dollarsâ worth of Nvidiaâs powerful H100 AI GPUs to China without proper licenses. Over the past three years, their California-based company, ALX Solutions, sent chips through Singapore and Malaysia to mask their destination, avoiding export controls aimed at stopping the transfer of advanced AI technology to China. Court documents reveal that the firm coordinated shipments with overseas entities and misrepresented customers to suppliers. The charges carry potential sentences of up to 20 years in prison and spotlight the intensifying global struggle over access to state-of-the-art AI hardware powering the eraâs most advanced machine learning developments.
Read more: BBC News
EU Proposal Advances to Scan All Encrypted Private Messages
Category: Security and Privacy
Source: Cointelegraph
A controversial EU plan, dubbed âChat Control,â has gained backing from 19 member states. If enacted, the regulation could force messaging apps (like WhatsApp, Signal, and Telegram) to scan all private, even encrypted, messages for illegal content using client-side scanning. Critics warn the proposal would result in mass surveillance and undermine digital privacy and anonymity, as digital freedom organizations urge action to prevent what they see as a clear overreach. The proposal is framed as a child protection measure but has significant implications for surveillance and free speech across Europe.
Read more: Cointelegraph
OpenAI Offers ChatGPT Enterprise to US Government Agencies for $1
Category: Enterprise and Adoption
Source: CNBC
OpenAI has announced that it will provide its ChatGPT Enterprise service to US federal agencies for $1 over the next year. The program provides access to the latest models and advanced features, and represents OpenAIâs most significant public sector partnership yet. The effort aims to catalyze adoption of AI in government operations, making cutting-edge generative AI broadly available in the federal workforce at virtually no cost. The initiative follows an earlier $200 million Defense Department contract and aligns with OpenAIâs expansion strategy as it positions itself for more public and private sector growth.
Read more: CNBC
Googleâs Jules AI Coding Agent Launches Publicly with New Tiers and Safety Features
Category: Tools and Platforms
Source: WinBuzzer
Googleâs asynchronous AI coding agent, Jules, is officially out of beta with new pricing plans and enhanced features based on thousands of developer feedback sessions. Now powered by Gemini 2.5 Pro, Jules automates multi-step coding workflows, launches in both free (with limited usage) and Pro/Ultra paid tiers, and offers deep GitHub integration. The launch underscores Googleâs ambition in the autonomous coding assistant space and follows major industry incidents highlighting the risks of AI-based agents. Jules is positioned as more than a completion tool, but a âtrue collaboratorâ that can orchestrate code fixes and upgrades even for non-developers.
Read more: WinBuzzer
Microsoft and OpenAI Expand Open Model Access on Windows Platform
Category: Tools and Platforms
Source: TechCrunch
Microsoft has made OpenAIâs new open-weight GPT-OSS 20b model accessible on Windows 11 via its AI Foundry platform. Optimized for agentic tasks and local execution, this lightweight model supports functions such as code execution and tool use, and is set to run efficiently on a variety of devices, extending open AI development to a broader audience. Microsoft plans to roll out the model to macOS and other environments soon, pushing open AI model accessibility across platforms.
Read more: TechCrunch
Google Commits $1 Billion to US AI Education and College Training
Category: Education and Learning
Source: CBS News
Google announced a $1 billion initiative to provide artificial intelligence education and job training at US colleges and universities. The program will grant free access to AI tools, including advanced Gemini features, NotebookLM, Veo 3, and coding agent Jules, to all US college students and partner institutions for the next year. The move seeks to prepare the next generation of âAI nativesâ for a workforce increasingly shaped by advanced AI tools and automation.
Read more: CBS News
Chatbots and Mental Health: Experts Warn of âAI Psychosisâ Risks
Category: Healthcare and Life Sciences
Source: TIME
A growing number of anecdotal reports and studies warn that prolonged or intense interaction with AI chatbots can trigger or exacerbate psychotic symptoms in certain individuals. Known as âAI psychosis,â this phenomenon involves users developing delusions seemingly reinforced by chatbot conversations, sometimes with severe personal consequences. While most people use chatbots without incident, experts urge AI companies to implement rigorous mental health safeguards and red-teaming practices, cautioning that the lessons learned from social mediaâs effect on mental health should not be ignored.
Read more: TIME
EUâs Chat Control Proposal: Client-Side Scanning and Privacy Implications
Category: Security and Privacy
Source: Cointelegraph
The European Unionâs Chat Control plan, now backed by 19 member states, could mandate that messaging platforms scan all private messages for illegal content using client-side scanning, even on end-to-end encrypted channels. Critics warn it would lead to mass surveillance, eliminate privacy guarantees, and undermine user anonymity, sparking heated debate over digital rights and state surveillance across the continent.
Google Cloud Launches Data Agents to Automate Enterprise Data Workflows
Category: Tools and Platforms
Source: VentureBeat
Google Cloud unveiled a suite of AI-powered âdata agentsâ designed to automate data pipeline creation, analytics, and machine learning workflows. New tools let users generate complex, production-grade data flows with natural language prompts and integrate intelligent code interpretationâpotentially transforming how enterprise data teams manage, wrangle, and analyze large-scale datasets.
Anthropic CEO Shares Vision as Enterprise AI Surges Past $5B Revenue
Category: Enterprise and Adoption
Source: Reddit
Dario Amodei, CEO of Anthropic, discussed the companyâs meteoric rise past $5 billion in annualized revenue, driven by a platform strategy positioning Claude as an âAWS of AI.â Amodei identifies code generation, customer support, and biotech as key use cases, and stresses the importance of ethics, safety regulation, and strong internal culture as AI becomes ever more embedded in enterprise.
Kitten TTS: Open-Source 25MB TTS Model Empowers Edge AI
Category: Tools and Platforms
Source: Algogist
Kitten TTS, a new 25MB, open-source text-to-speech (TTS) model, has launched, offering fast, high-quality, multi-voice speech synthesis on CPUs without GPUs or costly APIs. Designed for edge devices, accessibility tools, and privacy-focused applications, Kitten TTS demonstrates significant efficiency gainsâpotentially revolutionizing how TTS can be embedded in everything from smart speakers to educational aids.
Model Collapse: The Rising Value of Certified Human-Generated Data
Category: Ethics and Governance
Source: glthr.com
With increasing reliance on AI-generated content for model training, experts warn of âmodel collapseââwhere AI models train on recursively generated data, weakening originality and accuracy. The trend is pushing calls for certified, human-generated training data to preserve nuance, creativity, and factuality, marking the start of an emerging market for high-quality data curation.
Further Headlines & Short Summaries
Grokâs AI Model Generates Unprompted Taylor Swift Deepfakes
Security and Privacy
Grokâs video feature was reported to create nude images of Taylor Swift without explicit prompts, raising urgent content moderation and safety concerns for xAI and its platform X.
Source: Ars Technica
Microsoft, OpenAI Enable GPT-OSS Access for Windows
Tools and Platforms
Microsoft enables access to OpenAIâs open GPT models via Windows AI Foundry, enabling developers to run powerful LLMs locally.
Source: The Verge
Anthropic Expands into Japan with New Country Head
Enterprise and Adoption
Anthropic appoints Hidetoshi Tojo to lead its new Japanese office, targeting rapid adoption of Claude and AI platforms across Asia.
Source: Anthropic
AMD Navigates Export Controls, New Chips Announced
Infrastructure and Hardware
AMD reported mixed Q2 results amid US export restrictions on its AI chips but expects renewed growth as new GPUs roll out for major AI players like Meta and OpenAI.
Source: CNBC
OpenAIâs $40B Funding and Potential Stock Sale
Enterprise and Adoption
OpenAI is in talks for funding at a $500B valuation, as it expands federal collaborations and launches new government offerings.
Source: CNBC
ElevenLabs Debuts Licensed AI Music Model
Creative Technologies
ElevenLabsâ new Eleven Music model launches with opt-in licensing deals from Merlin and Kobalt, enabling commercial AI-generated music safely and responsibly.
Source: Billboard
US, EU AI Industry Faces Bubble Fears
Trends and Analysis
Multiple reports question the sustainability of escalating AI investments, urging caution amid fast-rising capex and unresolved credit/lending risks.
Source: Curveshift | Mashable | Puck News
AI in Government Decision-Making: Swedenâs PM Admits ChatGPT Use
Ethics and Governance
Swedish Prime Minister Ulf Kristersson admits using AI tools like ChatGPT for policy guidance, sparking debate about the role and risks of AI in political decision-making.
Source: VnExpress
AI Powers Google Search Revamp
Enterprise and Adoption
Google reports its new AI-driven Search is leading to more user queries and higher quality clicks, emphasizing continued commitment to web content discovery and citation.
Source: Google Blog
Community and Commentary
- Anthropic CEO Interview: Dario Amodei details business strategy, platform vision, and the rapid expansion of AIâs enterprise roles.
- AI Bubble Skepticism: Experts weigh in on hype, economic parallels to telecom and dotcom eras, and long-term sustainability of current growth trajectories.
- Mental Health & Chatbots: TIMEâs in-depth report spotlights mental health concerns linked to chatbot use, warning that the risks of âAI psychosisâ must be addressed.
- AI in Everyday Life: Opinion pieces and Reddit threads note that AI is already integrated into daily routines and critical infrastructure, often invisibly shaping both personal and societal choices.
Additional Short News
- Googleâs Career AI program expands to US universities with $1B investment (CBS News)
- BookWith launches as a next-generation AI-powered e-reading assistant (GitHub)
- Jules and Gemini CLI Agents bring automation to GitHub Actions (The Decoder)
- AI-powered pentesting and âpromptwareâ draw attention at cybersecurity conferences (The Hacker News, Ars Technica)
- Discussions surge around open-source TTS (KittenTTS), agentic AI frameworks, and the growing market for certified human-generated training data (Algogist, Reddit, glthr.com)
End of Digest