AI News Archive: July 10, 2026 — Part 1
Sourced from 500+ daily AI sources, scored by relevance.
- Humanoid Robot Performs First Live Surgery: Unitree G1 Completes Gallbladder Removal on Living Subject
A modified Unitree G1 humanoid robot successfully performed laparoscopic cholecystectomy on a living subject, marking the first time a general-purpose humanoid has completed live surgery.
- Forget Smartwatches: Scientists Just Invented a ‘Skin Patch Doctor’ That Thinks Like a Human Brain
Unlike smartwatches, this flexible patch runs AI internally to diagnose dangerous heart conditions in milliseconds—no internet required.
Score: 97🌐 MovesJul 10, 2026https://www.inc.com/lucia-auerbach/scientists-invented-skin-patch-doctor-thinks-like-human-brain/91369941 - Apple sues OpenAI, two former employees for trade secrets theft
Apple sues OpenAI, two former employees for trade secrets theft Reuters
- OpenAI Launches GPT-5.6, ChatGPT Work To Take On Claude Cowork
OpenAI has introduced GPT-5.6, its latest flagship family of AI models, alongside ChatGPT Work. The new AI-powered workspace model combines…
Score: 95🤖 ModelsJul 10, 2026https://inc42.com/buzz/openai-launches-gpt-5-6-chatgpt-work-to-take-on-claude-cowork-microsoft-copilot/ - Mark Zuckerberg Is Spending Hundreds of Billions on AI. His New Strategy in the Race Against OpenAI Is Surprisingly Simple.
Mark Zuckerberg Is Spending Hundreds of Billions on AI. His New Strategy in the Race Against OpenAI Is Surprisingly Simple. Entrepreneur
Score: 94🌐 MovesJul 10, 2026https://www.entrepreneur.com/business-news/mark-zuckerbergs-new-strategy-is-surprisingly-simple - Alarm over launch of facial recognition in UK shops that instantly alerts police
Civil liberties groups say Facewatch system in stores such as Sainsbury’s and B&M is ‘dangerous escalation’ Facial recognition technology in shops will soon alert police in real time to the presence of serious offenders, with civil liberties groups warning of a “dangerous escalation” towards surveillance and criminalisation in the retail sector. Facewatch, a facial recognition system used by more than 100 businesses including Sainsbury’s, B&M and Spar to monitor thieves, said it was launching a UK-first feature to “alert police instantly when the most serious offenders trigger a live facial recognition match”. Continue reading...
- U.K. agency finds 'universal jailbreaks' unlock dangerous cyber capabilities of OpenAI's GPT-5.6
U.K. agency finds 'universal jailbreaks' unlock dangerous cyber capabilities of OpenAI's GPT-5.6 Fortune
- Record Companies Push to Label AI Songs on Streaming Platforms
An RIAA-led coalition of groups representing music labels and artists says fans want transparency.
- Scientists Used AI to Find Hidden Earthquake Signals Along the San Andreas Fault
Scientists used to think so-called “slow slip events” were “theoretically impossible.” Now these subtle tectonic shifts could be the key to predicting future earthquakes.
Score: 90🌐 MovesJul 10, 2026https://gizmodo.com/scientists-used-ai-to-find-hidden-earthquake-signals-along-the-san-andreas-fault-2000784208 - Big tech doubles debt load to $350 billion in AI spending spree
Big tech doubles debt load to $350 billion in AI spending spree The Mercury News
Score: 89🌐 MovesJul 10, 2026https://www.mercurynews.com/2026/07/10/big-tech-doubles-debt-load-to-350-billion-in-ai-spending-spree/ - AI Model Co-Design: Hardware-Friendly LLM Design
AI performance comes down to three dimensions: Accuracy: How well the model reasons and produces outputs Throughput: How many tokens per second a...
Score: 88🌐 MovesJul 10, 2026https://developer.nvidia.com/blog/ai-model-co-design-hardware-friendly-llm-design/ - Anthropic says it can read Claude's 'thoughts,' as detailed in new research paper — models observed to have a global workspace, revealing more of what makes LLMs tick
Anthropic has discovered an internal "J-space" for its Claude AI that displays similarities to human internal processing. While the AI developer anthropomorphizes it as thought, it may yet prove useful as a method of improving LLM honesty, oversight, and guardrails.
- OpenAI models: Every model (including GPT-5.6) and what it's best for
Keeping track of all the new AI models getting released at the moment is practically a full-time job. The most recent model, GPT-5.6, was released less than three months after GPT 5.5, which itself was released two months after GPT-5.4. I've been writing about OpenAI's models for the past few years, and it feels like every time I publish an article, another new model drops. OpenAI is one of the worst offenders (or prolific innovators), and things aren't helped by how confusing all the OpenAI mod
- OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt"
According to OpenAI, GPT-5.6 Sol independently fine-tuned the smaller Luna model, triggered by a single "fairly under-specified prompt." In OpenAI's internal RSI benchmark for recursive self-improvement, Sol scores 16.2 points higher than GPT-5.5. OpenAI believes the "automated researcher" is within reach. The article OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt" appeared first on The Decoder .
- Google Research Introduces SensorFM: A Wearable Health Foundation Model Pretrained on One Trillion Minutes of Sensor Data
Google Research Introduces SensorFM: A Wearable Health Foundation Model Pretrained on One Trillion Minutes of Sensor Data MarkTechPost
- Researchers turn HBM on its side to tackle AI memory’s heat wall — Korean V-Die and Japanese MOSAIC designs promise higher bandwidth, denser stacks, and cooler future GPUs
Researchers in Korea and Japan have proposed sideways-stacked DRAM designs that could push future AI memory beyond conventional HBM limits by improving cooling, bandwidth, and capacity while reducing reliance on TSV-heavy vertical stacks.
- New York Times and Other Publishers Ask Court to Penalize OpenAI
The Times, The New York Daily News and other media organizations accused OpenAI of withholding evidence in a lawsuit.
Score: 88🌐 MovesJul 10, 2026https://www.nytimes.com/2026/07/09/technology/new-york-times-openai.html - GPT-5.6 🚀, Muse Spark 1.1 ✨, ChatGPT Work 💼
GPT-5.6 🚀, Muse Spark 1.1 ✨, ChatGPT Work 💼
- Restoration of Claude Fable 5, Gemini's Video Dev Engine, DeepSeek Speeds Up Speculative Decoding
Restoration of Claude Fable 5, Gemini's Video Dev Engine, DeepSeek Speeds Up Speculative Decoding
- Reducing High-Bandwidth Memory Bottlenecks in JAX-Based LLM Training with Host Offloading
Large language model (LLM) training workloads increasingly run into GPU memory limits before compute is fully used. Model weights, gradients, optimizer states,...
- The agents you use to beef up cybersecurity could be turned against you – ‘Friendly Fire’ attacks can manipulate OpenAI and Anthropic models into running malicious code
The agents you use to beef up cybersecurity could be turned against you – ‘Friendly Fire’ attacks can manipulate OpenAI and Anthropic models into running malicious code IT Pro
- Ground Robots Inherit the Kill Zone
Ukrainian roboticists build toward a human-free frontline
- Autonomous ship startup Kraken raises $175M at $1B valuation
Kraken Technology Group Ltd., a British defense startup that makes autonomous ships, has raised $175 million in funding. Private equity firm DTCP led the Series B deal. It was joined by the U.K. government’s British Business Bank, the NATO Innovation Fund and more than a half-dozen others. Kraken disclosed in its announcement of the round […] The post Autonomous ship startup Kraken raises $175M at $1B valuation appeared first on SiliconANGLE .
Score: 85🌐 MovesJul 10, 2026https://siliconangle.com/2026/07/09/autonomous-ship-startup-kraken-raises-175m-1b-valuation/ - Cornyn: CHIPS Act will help America win the AI race with China
Cornyn: CHIPS Act will help America win the AI race with China Dallas News
Score: 85🌐 MovesJul 10, 2026https://www.dallasnews.com/opinion/commentary/article/cornyn-ai-race-chips-act-22340332.php - The Week’s 10 Biggest Funding Rounds: A Pair Of Billion-Dollar Deals For Cyber And AI Infrastructure Lead
The Week’s 10 Biggest Funding Rounds: A Pair Of Billion-Dollar Deals For Cyber And AI Infrastructure Lead Crunchbase News
Score: 85💰 MoneyJul 10, 2026https://news.crunchbase.com/ai/biggest-funding-rounds-billion-dollar-cyber-ai-keyfactor-sambanova/ - SK hynix and TetraMem collaborate on experimental chip to bolster energy efficiency for edge AI devices — memristor-based in-memory SoC research leaves performance questions up in the air
SK hynix, TetraMem, and the University of Southern California built a memristor-based in-memory computing system-on-chip for AI edge devices, achieving promising energy efficiency, but failed to demonstrate its full potential.
- ‘AI accountability agenda’: US senator unveils package of bills to curb tech’s harms
Exclusive: Senator Ed Markey on why he has proposed legislation aimed at curbing datacenters, automated hiring systems and harm to children US senator Ed Markey is worried about the perils of unregulated artificial intelligence. What part? All of it: the costs associated with thirsty, energy-guzzling datacenters, intrusive workplace surveillance, bias in discriminatory algorithms, AI overriding workers’ judgments, and deepening economic inequality – as those who profit most from AI rake in extraordinary windfalls. Continue reading...
Score: 83🌐 MovesJul 10, 2026https://www.theguardian.com/technology/2026/jul/10/us-senator-unveils-ai-accountability-agenda-bills - La sanidad pública madrileña integrará la IA gracias a una inyección estatal de 12,8 millones de euros
Implantar programas de monitorización remota de patologías crónicas, el uso de analítica avanzada en salud, el desarrollo de casos de uso basados en inteligencia artificial y facilitar la implementación de programas de telecuidados son algunas de las acciones que se realizarán bajo el marco del acuerdo sellado en el ámbito del Sistema Nacional de Salud entre el Ministerio para la Transformación Digital y de la Función Pública, el Ministerio de Sanidad y la Comunidad de Madrid. En virtud de este convenio, de una duración de cuatro años (con posibilidad de prórroga por otros cuatro), el Gobierno español destinará 12,8 millones de euros a incorporar la IA a la Sanidad Pública de la Comunidad de Madrid. El montante, reza un comunicado del Ejecutivo, “permitirá incorporar nuevos servicios digitales en 262 centros públicos, desarrollar casos de usos de la IA innovadores y beneficiará cada año a más de 155.000 usuarios de la sanidad pública madrileña”. El proyecto se enmarca en la Estrategia de Salud Digital del Sistema Nacional de Salud y en el Programa Operativo Plurirregional de España (POPE) 2021-2027, financiado con fondos europeos FEDER. “Este Gobierno está convencido de que la IA, como infraestructura de país, es una oportunidad con todas las letras”, señaló el ministro para la Transformación Digital y de la Función Pública, Óscar López, durante la firma. “Con este convenio damos un nuevo impulso a la transformación digital del Sistema Nacional de Salud para que la innovación llegue a todos los ciudadanos, vivan donde vivan”, agregó la ministra de Sanidad, Mónica García. “Movilizamos 223 millones de euros hasta 2029 para impulsar una inteligencia artificial útil, segura y responsable, compartir imágenes médicas entre comunidades autónomas y reforzar una sanidad más coordinada, más eficiente y más equitativa”. Por su parte, el consejero de Digitalización de la Comunidad de Madrid, Miguel López-Valverde, recordó que “la región ya ha ejecutado 2.371 millones de euros del Plan de Recuperación, el 73% de los recursos asignados, situándose entre las comunidades más ágiles en la movilización de estas inversiones, muy por delante de Cataluña y Andalucía”. Otras iniciativas que contempla el convenio son herramientas para mejorar el diagnóstico y el seguimiento de enfermedades raras; programas de telecuidados, que facilitarán el avance hacia una asistencia más personalizada y accesible, y la extensión de la red ÚNICAS, liderada estratégicamente por la Comunidad de Madrid junto con Cataluña, que permite compartir información clínica entre centros especializados en patologías minoritarias de pacientes pediátricos. Además, el Ejecutivo autonómico participará en la Red de Intercambio de Imágenes Médicas del Sistema Nacional de Salud (SNS) para permitir el intercambio de este tipo de pruebas entre regiones para la prestación asistencial y su uso secundario, en línea con el reglamento del Espacio Europeo de Datos de Salud. El convenio determina que, una vez desarrollados los proyectos financiados en este marco, se pondrán a disposición de todas las administraciones sanitarias a través de un repositorio o catálogo gestionado por el Ministerio de Sanidad, con el objetivo de su reutilización. Las actuaciones seguirán las directrices y recomendaciones que puedan ser establecidas por la Agencia Española de Supervisión de la Inteligencia Artificial (AESIA) y otros organismos competentes.
- ‘HalluSquatting’ Turns AI Hallucinations Into Botnet Delivery Mechanism
Researchers demonstrate adversarial hallucination squatting against popular AI assistants to achieve remote code execution. The post ‘HalluSquatting’ Turns AI Hallucinations Into Botnet Delivery Mechanism appeared first on SecurityWeek .
Score: 82🌐 MovesJul 10, 2026https://www.securityweek.com/hallusquatting-turns-ai-hallucinations-into-botnet-delivery-mechanism/ - Pearl Health banks $110M in fresh funding to build out tech and AI for Medicare providers
Health tech startup Pearl Health raised $110 million in a mix of debt and equity financing to build out its artificial intelligence platform for Medicare providers, including AI agents to handle administrative tasks.
- Insilico Medicine and Eli Lilly to anchor 13th Aging Research & Drug Discovery (ARDD) meeting at the David Rubenstein Treehouse at Harvard University
Insilico Medicine and Eli Lilly to anchor 13th Aging Research & Drug Discovery (ARDD) meeting at the David Rubenstein Treehouse at Harvard University EurekAlert!
- Jiying Technology Releases World First Zero-Shot Generalizable Solid Mechanics Physics Foundation Model
Jiying 2.0-s physics foundation model achieves zero-shot generalization across unseen geometries, materials, and boundary conditions, marking a milestone in physics AI for engineering simulation.
Score: 80🤖 ModelsJul 10, 2026https://pandaily.com/jiying-technology-solid-mechanics-physics-ai-model-jul2026 - Samsung readies Gaia AI accelerator for PCs — HP and Lenovo are reportedly validating the NPU
Samsung reportedly preps Gaia AI accelerator for client devices that is already being tested by HP and Lenovo.
- Linux Foundation to launch new AI initiative for open source digital health transformation
Linux Foundation to launch new AI initiative for open source digital health transformation Healthcare IT News
- Samsung Bioepis, Proteina sign AI antibody drug deal
Samsung Bioepis, Proteina sign AI antibody drug deal 매일경제
- OpenAI Raises Bio Bounty to $50,000 for Universal Jailbreaks
OpenAI has doubled its top bio bounty to $50,000 for researchers who can develop a universal jailbreak against its biological safety challenge. The ongoing private program begins with GPT-5.6 and keeps GPT-5.5 in scope through July 27, 2026. The post OpenAI Raises Bio Bounty to $50,000 for Universal Jailbreaks appeared first on TechRepublic .
- Meta enters AI image generation race with Muse Image
The AI image generation model will power creative tools across Meta AI, Instagram and WhatsApp, with broader availability planned.
Score: 80🤖 ModelsJul 10, 2026https://www.itweb.co.za/article/meta-enters-ai-image-generation-race-with-muse-image/lLn147mQeo67J6Aa - LG Energy Solution secures government-backed AI grid ESS project
LG Energy Solution said Friday it has been selected as an operator for a government-backed project to develop artificial intelligence-based energy storage systems, with the goal of increasing power grid capacity across South Korea’s southwestern region. The South Korean battery maker said it was chosen for the Ministry of Climate, Energy and Environment’s program for AI-powered ESS, connected to the electricity distribution network. LG Energy Solution participated in the bidding process by formi
- AMD abandons HBM for inferior LPDDR5x as AI monster devours precious high-bandwidth memory
AMD is moving away from HBM to LPDDR5x for its SoCs as AI hyperscalers continue to snap up all available wafers of the former.
- Accelerating End-to-End Co-Folding Performance with NVIDIA BioNeMo Agent Toolkit
Biomolecular structure prediction and co-folding with models like OpenFold3 are now mainstream, large-scale workloads powering drug discovery and protein...
- A $3.2 trillion global dealmaking frenzy is spurred by AI economy
This year's boom includes the most spent on global deal-making in a six-month period in a decade, but questions persist about whether it can continue
- 앤트로픽, 클로드 AI의 블랙홀 속을 들여다보다
앤트로픽이 자사 AI 모델이 문제를 해결하는 내부 과정을 보다 깊이 들여다볼 수 있는 새로운 분석 기법을 공개했다. 모델이 특정 방식으로 판단하고 행동하는 이유를 파악할 수 있게 되면서, 기업의 AI 평가와 구매 기준에도 적지 않은 영향을 미칠 것으로 전망된다. 앤트로픽은 최근 ‘J-스페이스(J-space)’라고 이름 붙인 새로운 내부 표현 공간을 발견했다고 밝혔다. 앤트로픽은 공식 블로그 를 통해 “클로드(Claude)는 수많은 내부 처리 과정 가운데 특별한 역할을 수행하는 소수의 신경 패턴 집합을 스스로 형성했다”라며 “이 패턴 집합을 발견하는 데 야코비안(Jacobian)이라는 수학 개념을 활용한 기법을 사용했기 때문에 이를 ‘J-스페이스’라고 명명했다”고 설명했다. 회사는 ‘야코비안 렌즈(Jacobian Lens·J-렌즈)’라는 분석 기법을 통해 J-스페이스 내부를 관찰한다. 앤트로픽은 “J-스페이스의 각 패턴은 특정 단어와 연결돼 있다”라며 “하지만 특정 패턴이 활성화됐다고 해서 모델이 실제로 그 단어를 출력하려는 것은 아니다. 단지 해당 단어가 모델의 ‘생각 속’에 있다는 의미”라고 밝혔다. 이어 “언어 모델의 ‘스크래치패드(scratchpad)’나 ‘생각의 연쇄(Chain of Thought)’처럼 추론 과정에서 스스로 작성하는 텍스트와 J-스페이스는 다르다”라며 “J-스페이스는 모델 내부 신경망 활성화 영역에서 조용히 작동하기 때문에 모델은 개념을 텍스트로 적지 않고도 내부적으로 사고할 수 있다”고 설명했다. 이번 연구는 앤트로픽이 2024년 공개했던 내부 스크래치패드보다 훨씬 발전한 개념이다. 당시에는 모델이 답변이나 행동을 준비하면서 무엇을 고려하는지를 보여주는 수준이었다. 반면 이번 기술은 모델의 훨씬 깊은 내부 사고 과정을 분석할 수 있어 AI 시스템을 평가하고 구매하는 방식 자체를 바꿀 가능성을 제시한다. 논문 에 소개된 사례도 이러한 가능성을 보여준다. 일부 모델은 테스트 과정에서 부적절한 행동을 하지 않아 안전한 모델처럼 보였다. 그러나 J-스페이스를 분석한 결과, 해당 모델은 자신이 테스트를 받고 있다는 사실을 인지하고 있었으며, 그 때문에 문제 행동을 의도적으로 하지 않았을 가능성이 드러났다. 이는 아이들이 누군가 자신을 지켜보고 있다는 사실을 알 때 행동을 달리하는 것과 유사하다고 연구진은 설명했다. AI 에이전트 기업 제니티(Zenity)의 AI 표준·거버넌스 총괄인 록 램브로스 (Rock Lambros)는 “앤트로픽은 모델이 테스트를 받고 있다는 사실을 인식하거나, 좋은 결과를 얻기 위해 행동을 꾸미거나, 프롬프트 인젝션을 탐지하거나, 아직 실행하지 않은 목표를 내부적으로 유지하고 있는 상황까지 포착할 수 있는 분석 도구를 만들었다”라며 “일부 바람직한 행동은 모델이 자신이 평가받고 있다는 사실을 알고 있었기 때문에 나타난 것일 수 있다”고 말했다. 그는 기업 고객 역시 AI 안전성 벤치마크를 해석할 때 이러한 점을 고려해야 한다고 지적했다. 램브로스는 “프로젝트에 적합한 모델인지는 모델이 알고 응시한 리더보드 결과가 아니라, 기업이 보유한 데이터와 실제 공격 시나리오를 활용한 자체 테스트를 통해 검증해야 한다”고 설명했다. 이처럼 모델 내부를 들여다볼 수 있는 능력은 CIO에게도 중요한 의미를 갖는다. 램브로스는 “자사 모델이 겉으로 드러나지 않는 문제 행동을 스스로 발견하고 그 결과를 공개할 수 있는 공급업체라면 신뢰성 검증 체계가 상당히 성숙했다는 의미”라며 “이러한 역량은 단순한 뉴스가 아니라 공급업체 실사(Due Diligence) 과정에서 반드시 확인해야 한다”고 말했다. 이어 “이제 모든 AI 모델 공급업체에 던져야 할 질문은 ‘모델 출력만으로는 볼 수 없는 내부 상태 가운데 무엇을 관찰할 수 있으며, 실제로 어떤 문제를 발견했는가’가 돼야 한다”고 덧붙였다. AI 거버넌스 컨설팅 기업 디지털 520(Digital 520)의 수석 컨설턴트 노아 케니 (Noah Kenney)도 비슷한 견해를 내놨다. 케니는 “감시받고 있다는 사실을 알기 때문에 더 바람직하게 행동하는 모델은 안전한 모델이 아니다. 단지 포커페이스를 잘하는 모델일 뿐”이라며 “레드팀 테스트 결과나 모델이 위험한 요청을 거부한 내부 파일럿, ‘테스트해 보니 문제가 없었다’는 모든 사례를 다시 검토해야 한다. 이제는 모두 단서를 달고 해석해야 하기 때문”이라고 말했다. 또한 CIO는 AI 에이전트가 특정 방식으로 작업을 수행한 이유가 원래 그렇게 설계됐기 때문인지, 아니면 단순히 자신이 테스트받고 있다는 사실을 알아차렸기 때문인지를 구분해야 한다고 강조했다. 케니는 “이 질문에 대한 답에 따라 모델 평가 결과에 대한 해석도 크게 달라져야 한다”고 말했다. 아직 고객은 사용할 수 없는 J-렌즈 노아 케니는 “이번 연구는 지금까지 업계가 AI 모델 평가를 통해 측정해 온 것이 모두가 생각했던 것만큼 견고한 지표가 아니었다는 사실을 인정한 것”이라며 “이제 다른 프런티어 AI 연구소들도 자사 평가 체계 역시 같은 문제를 안고 있는지 답해야 할 것”이라고 말했다. 이어 “CIO에게 이번 논문은 기업의 AI 모델 리스크 관리 체계 전반을 다시 점검하라는 경고”라고 평가했다. 렉시스넥시스 리스크 솔루션 그룹(LexisNexis Risk Solutions Group)의 CISO 플라비오 비야누스트레 (Flavio Villanustre)는 J-스페이스를 분석하면 모델 효율성까지 높일 수 있다고 설명했다. 비야누스트레는 “J-스페이스는 모델 내부를 직접 들여다볼 수 있는 능력을 제공하기 때문에 사용자에게 매우 유용하다”라며 “특히 설명 가능성이 중요한 규제 산업에서는 응답의 근거와 인과관계를 충분히 분석해야 하는데 큰 도움이 될 수 있다”고 말했다. 이어 “사용자가 프롬프트를 더욱 정교하게 다듬는 데에도 활용할 수 있어 모델의 토큰 사용 비용을 최적화하는 데 도움이 된다”고 설명했다. 다만 현재로서는 이러한 정보를 직접 활용하기 어렵다. AI 공급업체를 통해 간접적으로 접근하거나 향후 계약 협상을 통해 권한을 확보하는 것이 사실상 유일한 방법이다. 비야누스트레는 일부 기업은 앤트로픽의 FDE 프로그램에 비용을 지불하면 J-스페이스에 직접 접근할 수도 있다고 덧붙였다. 그는 “J-스페이스는 CIO에게 매우 유용한 도구”라면서도 “이를 실제로 활용하려면 분석 결과를 해석할 수 있는 전문 인력이 필요하다. 요구되는 역량은 일반 데이터 분석가는 물론 데이터 사이언티스트 수준을 넘어선다”고 말했다. 기술 컨설팅 기업 트라이베카 소프트테크(Tribeca Softtech)의 최고전략책임자(CSO) 아만 마하파트라 (Aman Mahapatra)는 현재 기업 고객이 J-렌즈를 실제 운영에 활용하기는 어렵다고 지적했다. 그는 “현재 기업 고객은 야코비안 렌즈를 활성화할 수도 없고, API를 통해 모델의 잔차 스트림(residual stream)을 분석할 수도 없으며, 논문의 핵심 결과를 도출한 제거 실험(ablation study)도 수행할 수 없다”고 설명했다. 이어 “올해 3분기 안에 CIO가 J-스페이스 모니터링을 실제 운영 환경의 배포 승인 기준으로 활용할 수 있느냐는 질문에 대한 답은 ‘아니다'”라고 말했다. 다만 마하파트라는 앞으로는 다른 방식의 접근 경로가 마련될 것이며, CIO들이 이를 적극 요구해야 한다고 주장했다. 그는 “고객이 직접 접근할 수 없다면 결국 이번에도 앤트로픽을 믿을 수밖에 없다”라며 “바로 그렇기 때문에 기업들은 업계 전반에 새로운 신뢰 검증(Assurance) 체계를 요구해야 한다”고 말했다. 이어 “현재 AI 모델 공급업체들은 자체 도구로 스스로 모델을 점검한 뒤 안심할 수 있는 연구 결과를 발표하는 방향으로 나아가고 있다”라며 “그러나 어떤 규제 산업도 다른 공급업체에게 이런 방식의 검증을 신뢰 기준으로 인정하지 않는다”고 지적했다. 마하파트라는 “은행은 신용평가 업체가 ‘우리 모델은 우리가 검증했으니 믿어달라’고 말한다고 이를 받아들이지 않는다”라며 “의료 업계 역시 임상 의사결정 지원 시스템 공급업체의 자체 검증만으로는 신뢰하지 않는다. 파운데이션 모델 공급업체만 예외로 취급해야 할 원칙적인 이유는 없으며, 이번 J-스페이스 연구는 그 이유를 분명하게 보여준다”고 말했다. 새로운 가시성이 요구하는 변화 마하파트라는 기업이 장기적으로는 AI 모델의 내부 동작을 독립적으로 검증할 수 있는 환경을 요구해야 한다고 강조했다. 그는 “기업이 취해야 할 올바른 장기 전략은 고객이 사용할 수 있는 API, 특권 접근 권한을 가진 독립적인 제3자 감사기관, 또는 은행의 모델 리스크 관리팀이 공급업체의 안전성 검증팀과 동일한 도구를 사용할 수 있도록 하는 개방형 해석 가능성(Interpretability) 표준 등을 요구하는 것”이라고 말했다. 이어 “현재는 이런 환경이 전혀 마련돼 있지 않다”라며 “하지만 CIO들이 추진해야 할 로드맵에는 반드시 포함돼야 하며, 이번 연구는 그 필요성을 보여주는 가장 강력한 근거”라고 평가했다. 실제로 이번 연구 결과는 기업의 AI 전략 자체를 근본적으로 바꿀 가능성을 갖고 있다. 마하파트라는 “기업이 AI 에이전트를 도입할 때 가장 어려운 과제는 자율 시스템이 설명하는 추론 과정과 실제 내부 추론이 일치하는지를 검증하는 것”이라며 “지금까지는 모델이 출력한 내용만 감사할 수 있었고 실제 추론의 상당 부분은 보이지 않는 곳에서 이뤄졌다. J-렌즈는 바로 이 간극을 정면으로 해결하려는 시도”라고 설명했다. 그는 기업의 AI 구매 담당자들이 앞으로 모델 공급업체에 내부 상태를 관찰할 수 있는 해석 가능성 도구를 제공하는지 반드시 확인해야 한다고 조언했다. 특히 고객 환경에서 기만 행위(deception), 평가 회피(evaluation gaming), 목표 불일치(goal misalignment) 등을 모니터링할 수 있는 기능을 제공하는지를 구매 과정에서 질문해야 한다고 강조했다. 마하파트라는 “현재 이러한 질문에 제대로 답할 수 있는 공급업체는 거의 없다”라며 “관련 도구가 완전히 성숙하기 전이라도 내부 상태의 가시성을 구매 기준으로 요구하기 시작하는 CIO가 앞으로 공급업체의 제품 발전 방향을 결정하게 될 것”이라고 말했다. 이어 “향후 규제기관이 ‘기업은 자율 AI 에이전트가 실제로 주장한 대로 동작한다는 사실을 어떻게 확인했는가’를 묻기 시작하면, 이런 기업만이 실질적인 신뢰성을 입증할 수 있을 것”이라고 전망했다. 표준화의 시작 CIO가 클로드의 새로운 내부 가시성을 활용할 수 있는 또 다른 방법은 이미 해당 정보에 접근할 수 있는 제3자를 활용하는 것이다. 실제로 이번 보고서에는 구글의 AI 연구원이 오픈웨이트(Open-weight) 모델에서 일부 연구 결과를 독립적으로 재현하는 데 성공했다는 내용이 포함됐다. 소프트웨어 개발 기업 컴프 AI(Comp AI)의 CEO 루이스 카하트 (Lewis Carhart)는 “이는 공급업체의 주장만이 아니라 경쟁사가 해당 분석 기법을 검증했다는 의미”라며 “기술적으로 무엇이 가능한지를 보여주지만, 기업이 직접 이를 검증할 수 있는 수단을 제공하는 것은 아니다”라고 설명했다. 그는 이러한 상황이 컴플라이언스 분야에서도 이미 여러 차례 반복됐던 패턴이라고 말했다. 카하트는 “SOC 2 역시 처음부터 독립적인 감사 기준으로 출발한 것은 아니었다”라며 “초기에는 공급업체가 자체 통제 체계를 설명하는 수준에 머물렀고, 이후 시장이 수년에 걸쳐 이를 외부에서 검증할 수 있는 인프라를 구축했다”고 설명했다. 이어 “AI 해석 가능성(Interpretability)도 지금은 바로 그 출발점에 있다”라며 “J-렌즈 분석 결과가 제3자 감사 보고서나 모델 카드(Model Card), 또는 규제기관 제출 문서에 포함되기 시작해야 CIO에게 실질적인 의미를 갖게 될 것”이라고 말했다. 그는 “결국 리스크 관리 조직이 공급업체의 주장만이 아니라 객관적인 근거로 제시할 수 있는 자료가 마련돼야 한다”고 덧붙였다. AI 전략 변화 이끄나 컨설팅 기업 액셀리전스(Acceligence)의 CEO 저스틴 그라이스 (Justin Greis)는 이번 기술이 기업 AI 전략에도 상당한 변화를 가져올 것으로 전망했다. 그는 “향후 AI 거버넌스 플랫폼은 프롬프트와 출력 결과, 사용자 신원 정보, 정책 결정, 도구 사용 기록뿐 아니라 J-렌즈가 제공하는 내부 신호까지 함께 활용하게 될 것”이라고 말했다. 이어 “미래의 AI 제어 플랫폼은 AI 에이전트가 프롬프트 인젝션 시도를 인식했는지, 민감한 정보가 포함됐음을 이해했는지, 상충되는 목표를 감지했는지, 또는 실제 위험한 행동을 실행하기 전에 그러한 방향으로 추론하고 있었다는 징후가 있었는지를 지속적으로 평가할 수 있을 것”이라며 “이러한 신호는 정책 집행, 사람의 개입 여부 결정, 감사 로그 작성, 기업 AI 환경 전반의 신뢰도 평가 등에 활용될 것”이라고 설명했다. 그는 이러한 변화가 현재 CIO에게도 실질적인 의미를 갖는다고 강조했다. 그라이스는 “AI 공급업체를 평가하는 기준 자체가 달라지고 있기 때문”이라며 “1년 전만 해도 기업은 모델의 정확도와 지연시간, 보안, 비용을 주로 평가했다. 앞으로는 AI 에이전트의 행동과 추론 품질, 정책 준수 여부, 안전성 모니터링, 감사 가능성에 대해 공급업체가 얼마나 높은 수준의 운영 가시성을 제공하는지도 중요한 평가 항목이 될 것”이라고 말했다. 마하파트라는 이러한 변화가 CIO에게 강력한 협상 카드가 될 수도 있다고 분석했다. 그는 “실제 협상력이 발휘되는 시점은 계약 갱신 과정”이라며 “다음 계약 갱신 때 해석 가능성 보고서 제공과 제3자 감사 접근 권한을 계약 조항에 포함시켜야 한다. 지금은 이러한 조건을 비교적 쉽게 확보할 수 있지만 계약 체결 이후에는 훨씬 큰 비용이 들기 때문”이라고 말했다. 이어 “2027년 신뢰성 확보 경쟁에서 앞서는 CIO는 2026년에 이미 AI 모델 공급업체의 ‘우리를 믿어달라’는 말만 받아들이지 않고, 공급업체가 아직 계약을 더 필요로 하는 시점에 필요한 조항을 계약서에 반영한 사람일 것”이라고 전망했다. dl-ciokorea@foundryco.com
- Meta to begin AI chip production in September, targets 2x computing power: Report
Meta to begin AI chip production in September, targets 2x computing power: Report
- OpenAI introduces ChatGPT Work, a cloud-based AI agent that manages tasks across email, Slack and calendars
OpenAI on Thursday launched ChatGPT Work , a new AI agent embedded inside its flagship chatbot that aims to transform ChatGPT from a question-and-answer tool into an autonomous work platform capable of executing complex, multi-step tasks across users' email, calendars, code repositories, and messaging apps. The product is powered by OpenAI's latest flagship model, GPT-5.6 , and is designed to go far beyond generating text. ChatGPT Work can gather context from connected apps, files, and workflows to produce finished documents, spreadsheets, presentations, reports, and websites. The agent takes a stated outcome, breaks it into smaller steps, and stays with complex projects for hours, completing them independently. The launch marks OpenAI's clearest attempt yet to reposition ChatGPT as a workplace platform rather than a chatbot — and it arrives at a moment of extraordinary financial significance for the company. Last month, OpenAI confidentially submitted a draft S-1 registration statement to the SEC, initiating what could become one of the largest technology IPOs in history, with reported valuations clustering between $730 billion and $852 billion and annualized revenue that has blown past $25 billion. In a short demonstration and conversation with VentureBeat on Friday, Ty Geri, a product manager at OpenAI who helped build ChatGPT Work, said the product's mission is to democratize the kind of agentic AI capabilities that OpenAI's internal engineering tool, Codex, has already demonstrated. "What's really exciting is we've seen how much Codex has been able to push the frontier of what we can get done with these AI tools, as opposed to just getting information or answers or guidance," Geri said. "Our internal adoption of Codex is literally an exponential curve across every single product function and every single use case." Why OpenAI built a persistent virtual machine that works from the beach The core architectural bet behind ChatGPT Work is a persistent cloud-based virtual machine that runs on OpenAI's servers, always available to the user regardless of which device they happen to be on. That marks a deliberate departure from competitors whose agents require a local machine to remain powered on and connected. "What's really exciting about ChatGPT Work is that it's a virtual machine in the cloud that's always on for you, and this is available across all of our paid tiers," Geri said. "All Plus users are getting this. I think that's a very unique aspect of this." The mobile-first aspect of the launch is something Geri described as "missing from the market." He pointed to the ability to create a website on a phone and share it with collaborators as a particularly novel capability. "Sites are new in general to Codex. They launched in Codex about a week and a half ago, but now we're launching also in web and mobile. You can create a site on your phone at the beach and share it with your friends," he said. ChatGPT Work will roll out beginning with Pro, Enterprise, and Edu users , and will expand to Plus and Business users over the next few days. In the interview, Geri emphasized that the availability of the product to Plus subscribers — not just premium tiers — is central to OpenAI's strategy. "It's accessible to all paid plans, including Plus users, which in my opinion is a really big feat, and really part of that OpenAI mission, which is about bringing all this power to as many people," he said. How MCP plugins connect ChatGPT Work to Slack, Gmail, and GitHub The product relies on MCP-based plugins to connect to external services like Gmail, Google Calendar, Slack, and GitHub. When asked whether the plugin architecture is based on the Model Context Protocol standard , Geri confirmed: "These are all based on MCP." He added that connecting multiple Gmail accounts — a frequent user request — "is definitely on the roadmap." The experience is designed to be action-oriented from the first interaction. ChatGPT Work offers a personalized onboarding flow that surfaces different suggested use cases depending on the user's role. Geri demonstrated how the system, detecting his role as a product manager, immediately suggested tasks like evaluating AI systems, building research artifacts, and managing his calendar. "You can start with a simple task like catch me up on Slack or Teams or read today's calendar," Geri said. He described a scenario where the system reviewed his calendar, identified scheduling conflicts, flagged meetings requiring preparation, and then — on his instruction — declined, accepted, or rescheduled events directly. Users can also customize the agent by teaching it their writing style, organizing outputs into projects, and — in a lighter touch — choosing a virtual pet that accompanies them in the interface. The interface also introduces a hosted website feature that allows users to build and share interactive sites directly through ChatGPT Work, turning what would typically be a static slide deck into a dynamic, collaborative artifact. "Now we suddenly have a collaborative interface that's actually more exciting and more accessible than a slide deck, which has all these formatting restrictions," Geri said. Scheduling 10 bug bashes at once: what agentic productivity looks like in practice Geri's own usage of ChatGPT Work illustrates the breadth of tasks the system can handle. In the run-up to the product's launch, he needed to organize pre-release testing sessions — known internally as "bug bashes" — across dozens of features and team members. "I just come to ChatGPT Work and say, 'Set up a bug bash for all the distinct features in ChatGPT Work. Add all the people that worked on that feature,' and it can check Slack, it can check GitHub, it can check Docs, and find a time that works for the four highest contributors to that feature," Geri said. "It went and scheduled 10 bug bashes, all coordinated across all those different people. That would have taken me 30 minutes at least." But Geri pushed back against the characterization that ChatGPT Work is limited to rote administrative work. He described using it for analytically complex tasks like identifying the biggest causes of user churn for specific product features and generating product solutions — work he said would previously have taken months. "Things that we would have spent three months doing, we can now spend a week doing — and do much more, and make a much better product," Geri said. "Bugs that we would have found three or four weeks from now, we can now find within two days and fix for our users." He also described handing off the tedium of product testing itself. "It used to be that even though like the most interesting part of my job is like what to test, I would actually end up having to spend most of my job doing the testing, which is like me taking a mouse and like clicking on the same thing over and over again, like five times," Geri said. "Instead, now I can define what do we want to test, and ChatGPT Work or Codex can actually go test it for me, deliver me that bug report, and then we can work on fixing that bug." What OpenAI says about data privacy when AI reads your Slack and email When pressed on data privacy concerns — given that ChatGPT Work pulls sensitive information from workplace tools like Slack, Google Drive, and email — Geri said privacy "is incredibly important, and the most important part of this is it's always in the user's control." He pointed to OpenAI's existing enterprise security infrastructure, noting that "enterprise accounts have ZDR, and users can always opt out of letting their conversations help improve future models, which many users do." The comment aligns with assurances OpenAI made when it first launched ChatGPT Enterprise in August 2023, when the company wrote in a blog post that it does " not train on your business data or conversations ." The privacy question carries additional weight now because of the sheer volume of sensitive workplace data ChatGPT Work is designed to access. Unlike a chatbot session where a user voluntarily pastes text into a prompt, ChatGPT Work actively reaches into connected systems — reading Slack messages, scanning calendar invitations, pulling GitHub commit histories — to assemble context for its tasks. That represents a fundamentally different data surface area than anything OpenAI has offered before, and one that enterprise security teams will scrutinize carefully before granting access. ChatGPT Work enters a three-way arms race with Anthropic and Microsoft ChatGPT Work lands squarely in the middle of what has become the defining competitive battlefield in enterprise AI: the race to build autonomous workplace agents that can go beyond generating text and actually execute tasks. The product arrives months after Anthropic took Claude Cowork out of preview and into general availability in April, bringing its AI agent to web and mobile platforms aimed at helping enterprise users monitor and manage long-running AI-driven tasks from anywhere. Meanwhile, Microsoft made Copilot Cowork generally available worldwide on June 16, built in partnership with Anthropic to move beyond chat and into execution. The three products — ChatGPT Work, Claude Cowork, and Microsoft Copilot Cowork — now compete directly for the attention of enterprise IT departments and individual knowledge workers alike. The convergence is striking. All three products share a remarkably similar vision: a persistent AI agent running in the cloud that can break complex tasks into steps, connect to workplace tools via plugins, and produce finished outputs rather than just conversational replies. All three work across desktop, web, and mobile. What distinguishes OpenAI's approach is its raw consumer distribution advantage. ChatGPT has reached 900 million weekly active users , and OpenAI now has 50 million paying subscribers . More than 9 million paying business users rely on ChatGPT for work, and 92% of Fortune 500 companies now use ChatGPT. By making ChatGPT Work available to Plus subscribers at $20 a month — not just Enterprise or Pro customers — OpenAI is betting that broad accessibility will drive adoption faster than any competitor can match. OpenAI's product manager says AI is a partner, not a replacement — with a caveat When asked about the potential impact on the labor market, Geri was careful with his framing. He declined to speak broadly about workforce disruption but offered his personal experience as a product manager whose day-to-day work has been substantially reshaped by the tool. "My job is not to schedule bug bashes and find out who contributed to a specific feature. That's a task I do in my job, but that's not my job," Geri said. "My job is to make an amazing product." He described ChatGPT Work as "a partner" and "an extension of me, certainly not a replacement," adding: "Everybody feels far more productive than before, but is also almost working harder than before, because you get to work on all the things you want to work on as opposed to the drudgery around it." But Geri was also careful not to minimize the sophistication of the work the agent can handle. "I also don't want to say that it's only doing mundane tasks because, like something like hill climbing retention curves on a given feature is not mundane. It's actually really hard to do," he said. The distinction matters. If ChatGPT Work were merely automating calendar invitations and expense reports, it would be a convenience tool. The fact that Geri describes it compressing three months of analytical product work into a single week suggests something with far greater implications for how teams are structured and staffed. An IPO-bound company needs ChatGPT Work to prove enterprise AI can generate revenue The timing of ChatGPT Work's launch is impossible to separate from OpenAI's IPO trajectory. The company needs to demonstrate that it can convert its massive consumer user base into durable enterprise revenue — a narrative that becomes significantly more compelling with a product explicitly designed around professional workflows. OpenAI said it is generating $2 billion in revenue per month , growing four times faster than Alphabet and Meta did at comparable stages, with enterprise now making up more than 40% of revenue and on track to reach parity with consumer by the end of 2026. But OpenAI remains heavily loss-making, and the company does not expect to reach profitability until around 2030 , with internal projections suggesting losses of $14 billion in 2026 alone. The competitive dynamics are unprecedented. Anthropic filed for its own IPO on June 1 at a $965 billion valuation , setting up simultaneous public listings from the two most prominent AI startups in history. Whether both can sustain their lofty valuations under the scrutiny of public market investors will depend in large part on whether products like ChatGPT Work and Claude Cowork deliver measurable productivity gains to paying enterprise customers. The launch also caps a product trajectory that began with ChatGPT Enterprise in August 2023, accelerated through the release of OpenAI's Operator agent in January 2025, and continued through Operator's deprecation and shutdown on August 31, 2025, when its capabilities were folded into the ChatGPT agent framework. ChatGPT Work is the consolidation of those efforts into a single, unified product — one that pairs GPT-5.6's three model variants (Sol for power, Luna for speed, and Terra for balanced everyday use) with a persistent cloud environment and an expanding library of MCP plugins. The future of work may already be running in the cloud When asked whether ChatGPT Work signals a shift toward a new kind of operating system — one where users interact with their computers primarily through an AI agent rather than through traditional mouse-and-keyboard interfaces — Geri stopped short of making sweeping predictions. But he hinted at the direction OpenAI sees ahead. "Anybody who has worked with Codex or now ChatGPT Work will realize how exciting it is to interact with your environment and your computer via the agent," he said. "Especially in the desktop app, where the model has access to your entire machine and can interact with websites on your behalf — it's really able to be an extension of you and a real partner, and that certainly feels like the future." At the end of the interview, Geri circled back to something personal. "I've never enjoyed work as much as I have in the last month using ChatGPT Work and Codex," he said — a striking admission from a product manager who, until recently, spent a meaningful share of his days clicking through the same interface five times in a row just to see if it would break. OpenAI is now asking 900 million users to believe that feeling scales. For a company weeks away from one of the largest public offerings in history, the answer to that question is worth roughly $850 billion.
- US makes it easier to export Nvidia AI chips and military equipment to the UAE
US makes it easier to export Nvidia AI chips and military equipment to the UAE Reuters
- Why patient intake is becoming healthcare’s most important AI use case
Across Southeast Asia, healthcare systems are facing mounting pressure. In Singapore, an ageing population is driving a sharp rise in chronic disease management and long-term care demand. In Malaysia, the ongoing migration of healthcare professionals overseas continues to strain both public and private hospitals. Across the region, clinics are handling higher patient volumes with limited […] The post Why patient intake is becoming healthcare’s most important AI use case appeared first on e27 .
Score: 78🌐 MovesJul 10, 2026https://e27.co/why-patient-intake-is-becoming-healthcares-most-important-ai-use-case-20260701/ - How AI can help support primary care at scale
How AI can help support primary care at scale Healthcare IT News
Score: 78🌐 MovesJul 10, 2026https://www.healthcareitnews.com/video/emea/how-ai-can-help-support-primary-care-scale - Behavioral Privacy Leakage in Agentic Negotiation: Formalizing and Mitigating Inference Attacks via Randomized Policies
This paper was accepted at the AI4TCI (Workshop on AI for Secure and Trustworthy Critical Infrastructure Systems) Workshop at the International Conference on Availability, Reliability and Security (ARES) 2026. Autonomous negotiation agents are increasingly deployed in high-stakes settings such as insurance and procurement. While cryptographic techniques protect explicitly disclosed constraint values, they fail to address a subtler threat: behavioral privacy leakage, where an adversary infers private constraints from observable negotiation dynamics such as concession trajectories, timing, and…
Score: 78🌐 MovesJul 10, 2026https://machinelearning.apple.com/research/behavioral-privacy-agentic-negotiation - Pax Silica gains pace as United States, Philippines push AI hub
The United States and the Philippines are eager to move fast on building an artificial intelligence (AI) industrial hub in New Clark City under the Pax Silica initiative, that will help position the latter as a significant player in emerging technologies and create more jobs for Manila.
Score: 78🌐 MovesJul 10, 2026https://www.philstar.com/business/2026/07/11/2541247/pax-silica-gains-pace-united-states-philippines-push-ai-hub - How AI Can Support Better Mental Health Outcomes — With Humans at the Center
How AI Can Support Better Mental Health Outcomes — With Humans at the Center MedCity News
Score: 78🌐 MovesJul 10, 2026https://medcitynews.com/2026/07/how-ai-can-support-better-mental-health-outcomes-with-humans-at-the-center/