저자: 미상 · 외국인
요약 보기
9/29 AI 뉴스 종합: 로맨스 스캠에 AI 페르소나 활용 증가(FBI·Anthropic 보고서), 엔비디아 주도 Open Secure AI Alliance 출범, 미중 AI 칩 밀수 수사, OpenAI 의회 직원용 프롬프트 팩, Claude Sonnet 5.5 Bedrock 출시, AI 자율 R&D 가속 경고 보고서. 마이클 버리는 AI 버블 붕괴가 예상보다 빠를 수 있다며 마이크론·엔비디아 풋옵션, SOXX 풋옵션 보유를 언급. 온세미-시낵틱스 주식 합병(상기 주가 급락).
다듬은 글
Welcome to your roundup of artificial intelligence news for September 29, 2026. Today, Nvidia created a new security alliance for AI agents, OpenAI launched specialized rapid response tools for Congressional staff, and Amazon released the Claude Sonnet 5.5 model. News number one: Scammers are now deploying entire fake dating apps filled with AI personas to trick users, according to a September report by Anthropic .
The study details a China-based operation that created a network of more than 20 fraudulent dating apps, using more than 4,700 AI personas to interact with at least 25,000 people. These automated agents generated approximately 2.36 million messages in just 2 weeks. The FBI notes that this tactic represents a significant increase in complexity, as attackers now use AI to replicate voices and personalized targeting to appear more convincing.
This shift comes as losses from romance fraud rose 38% last year, with total losses approaching $1 billion. Notably, Americans over the age of 60 were the hardest hit, losing over $500 million to these scams. As these AI-powered operations grow in scale, federal agents warn that the line between human interaction and automated fraud is becoming increasingly difficult to distinguish for victims seeking online connections.
News number two: Nvidia has joined 36 other organizations to launch the Open Secure AI Alliance, a collaborative initiative aimed at creating open- source security standards for the entire AI agent stack. Rather than limiting the scope to model security, this initiative targets the surrounding infrastructure, including identity management, permissions, isolation, protective fencing, and assessment systems.
Nvidia emphasizes that because open source is at the heart of modern cloud computing, defenders need systems they can test, deploy, and run on their own controlled infrastructure. The alliance is clearly focused on fostering a multi- vendor ecosystem to prevent single points of failure in these agent- based workflows .
The initial list of partners includes major industry players including Microsoft, IBM, Cisco, Salesforce, Adobe, and the Linux Foundation. This move is in line with a broader industry trend of shifting focus from basic model training to the operational security of autonomous agents. By standardizing logs and security tools across these 37 diverse partners, the group intends to create a unified system for protecting enterprise software and cloud environments.
News number three. The United States and China are locked in a complex geopolitical rivalry over artificial intelligence, with President Donald Trump declaring that whoever wins the race for AI dominance will win the future. The main hot spot remains the flow of advanced equipment, as the US restricts exports of high-end AI chips to China from 2022.
Despite these measures, federal prosecutors are actively investigating the smuggling of Nvidia chips into China. In a high-profile case, the Justice Department charged two Shiyan and Wally Lao, co-founders of Supermicro Computer, with violating export restrictions by supplying servers containing Nvidia hardware assembled in the United States to Chinese customers.
Lao has pleaded not guilty, while Nvidia claims to be following strict compliance protocols. The tension also extends to cloud access. When the Chinese Moonshot AI lab launched its Kimi K3 model with a large programming language, a White House official alleged that the firm had illegally obtained Nvidia chips and gained remote access to them through Thailand.
As both countries prioritize speed over cooperation, a lack of trust continues to define the landscape of global AI development. News number four. OpenAI Academy has released a set of specialized ChatGPT question packs designed for US Congressional staff. These resources cover a wide range of legislative and district functions, including legislative briefings, hearing preparation, voter outreach, and communications.
The packages provide structured templates for key roles ranging from chiefs of staff and legislative directors to social workers, planners, and field representatives. OpenAI for Government emphasizes that these tools are designed to speed up the document drafting process for time-consuming tasks, such as creating questions for hearings or summarizing source materials.
The company clearly states that the requests are not intended to make policy decisions or replace human oversight. Instead, the documentation instructs users to enter their own office-approved materials, verify all details, and retain final ownership of all commitments and positions. Moving away from a one-size-fits-all approach, Open AI is trying to integrate its models directly into specific Capitol Hill workflows.
This deployment is in line with a broader industry trend where large AI labs are creating dedicated, closed resources to encourage the adoption of generative tools in government and public sector institutions. News number five. Claude Sonnet 5.5 is now available on Amazon Bedrock and the Claude platform on AWS , offering increased efficiency for coding and knowledge-based tasks.
This latest iteration delivers faster speeds and lower cost per task compared to its predecessor, Sonnet 5. By integrating with the existing AWS ecosystem, the model supports regional data storage and leverages proven tools like AWS Identity and Access Management, CloudTrail, and CloudWatch for auditing and monitoring.
For engineering teams, the model is designed to perform limited work such as bug fixing, SQL generation, and UI/UX testing with the ability to verify results against specific requirements. In addition to code, it generates more polished documents, including architectural diagrams, one-page documents, and summary slides.
Because usage costs are reflected directly in your AWS bill, the model is designed for continuous, large-scale workloads, such as continuous agent monitoring or IDE-based coding assistance. This release follows Sonnet's standard deployment pattern, seeking to balance high-performance throughput with the operational control and security infrastructure that enterprise users already manage in their AWS environments.
News number six. A new report from the University of Cambridge's Artificial Intelligence Science and Policy Program warns that the automation of AI research and development could trigger an intellectual explosion, compressing years of progress into a matter of months. Authors Samuel Hammond and Sam Manning argue that because AI systems now write most of their code within their own development pipelines, the shift to fully automated research and development is already underway.
The article, published on September 28, 2026, notes that while this shift could accelerate the benefits of artificial intelligence, it poses extraordinary risks, including the potential loss of human control over superhuman systems. The authors emphasize that such rapid growth in capabilities may outpace society's ability to adapt, potentially undermining existing mechanisms of power checks between states, corporations, and branches of government.
Given the high stakes and current uncertainty, the report urges policymakers to prioritize gaining immediate visibility into the automated research and development process. This analysis comes as the industry moves beyond the possibilities seen even a year ago, shifting its focus to systemic risks that arise from recursive, self-perpetuating development cycles.
News number seven: xAI has officially added its Grok 4.7 Frontier model to the Amazon Bedrock catalog, expanding the platform's capabilities for coding and working with complex knowledge. This model has a context window of 500,000 tokens and introduces four different levels of configurable reasoning effort, from low to high.
Developers can access the model through the Bedrock runtime endpoint using cross-region output profiles with full support for feedback APIs, chat completion, and responses. According to the launch announcement on September 21, 2026, xAI built this iteration on a larger base model specifically optimized for endurance rather than pure speed.
The training process used a longer reinforcement learning run, prioritizing tasks that required hours of computation. This design choice allows the model to more closely validate its own results during complex, long-running agent workflows. By integrating Grok 4.7, Amazon Bedrock now offers a specialized tool for multi-step problem solving, positioning it among other high-level reasoning models currently competing for enterprise coding and automation workloads.
News number eight. California State University is gathering feedback from the San Francisco State University community regarding the integration and impact of generative artificial intelligence on campus. Students, faculty, and staff have until Monday, October 19 to complete an anonymous survey that takes less than 10 minutes .
Participants can also simply skip any questions and enter a drawing for an iPad or a $25 gift card. Andrew Roderick, SFSU's assistant vice president for academic technology, emphasized that the survey is open to everyone, regardless of their personal position or level of knowledge about these tools.
This initiative is a follow-up to last year’s system-wide survey, which surveyed more than 94,000 participants and revealed widespread concerns about academic integrity, job security, and ethical use. Researchers are now seeking to track how these attitudes and behaviors have evolved over time. According to the university, the data collected will directly influence future AI policy decisions within the CSU system.
This is a direct attempt to capture how campus students are actually experiencing this rapid technological shift. News number nine. Investor Michael Berry is accelerating his bearish chart on the artificial intelligence sector, shifting his strategy toward put options to take advantage of a potential market correction.
Berry, known for his high-profile short position in the housing market during the 2007-2009 financial crisis, said in his Monday newsletter that he now expects the artificial intelligence bubble to burst faster than previously thought . To obtain more cost-effective leverage, Berry is moving away from outright short positions, citing extremely tight volatility measures like the VIX as the reason for the current availability of put options.
His new portfolio includes put options on Micron with an expiration in June and a strike price of $500, as well as put options on Nvidia with a double-digit strike price for the same month. Additionally, he replaced his short position in the SOXX iShares Semiconductor ETF with September 2027 put options at a strike price below $400.
While Berry noted that some of these adjustments were aimed at managing tax liabilities, his primary motivation remains his belief that the AI industry could change dramatically next summer. News #10. Infropic is facing criticism over its recent claim that its AI agents independently discovered new enzymes, dubbed “ arts,” with potential for biotechnology.
Computational biologist Mario Rodríguez Mestre of the University of Copenhagen says he and his colleagues have been studying these specific enzymes and molecules for 4 years. Mestre says that because his team has been regularly using Anthropic models to write manuscripts and code for the past 3 years, they have shared key findings about these enzymes with the company.
He suggests that the main question is whether the AI reasoned about its discovery, or was guided by previously published data from his team. Mestre expressed concern that his research could be used to train future versions of the models. In a statement released Sunday, Anthropic denied the claims, saying the company was unaware of any previous published work describing the art system.
Additionally, the firm stated that Claude had not been trained in custom transcripts, and that its molecular biology team had no connection to Mestre's research. News #11. Two of the main architects of modern artificial intelligence, Nobel laureate Jeffrey Hinton and Joshua Bengio, have joined senior executives at OpenAI and Infropic to warn governments about a potential intelligence explosion.
Their new report, titled “What if Automation of AI R&D Triggers an Intellectual Explosion,” argues that we are approaching what could be the most important technological development in history. The authors define this phenomenon as a sharp acceleration driven by artificial intelligence, where progress that would normally take years is compressed to months or less.
Along with Hinton and Bengio, the paper was co-authored by more than 20 authors, including Anthropic co-founder Jack Clarke and OpenAI chief scientist Jakub Pachachi. The group calls on policymakers to take immediate action to reduce the risks of uncontrolled technological progress. This warning comes after a period of rapid growth in the industry, with large labs increasingly focusing on automating the research and development process itself.
The authors emphasize that the window for proactive management is narrowing as these systems gain the ability to repeat their own development cycles. News #12. On Semiconductor acquires Synaptics in a stock-for-stock deal to accelerate expansion of its physical AI business. The company expects this turnaround to increase its total addressable market by $30 billion , reaching a projected $243 billion by 2030.
CEO Hassan El- Khoury said the acquisition provides a strategic foundation for the company's move toward AI-focused computing platforms, robots and autonomous vehicles. The deal integrates Synaptics' Astro platform, which uses artificial intelligence processors and wireless connectivity to make real- time decisions on local hardware.
El-Khoury emphasized that the two product portfolios do not overlap, creating a complementary path for future research and development. The move comes after a period of market volatility for On Semiconductor, with shares suffering their steepest drop since March 2020 following the announcement. By shifting its focus to edge AI, the company aims to meet the demand for systems that perceive and process data independently of the cloud.
News #13. The International Olympic Committee has officially launched its first trusted artificial intelligence system to regulate the use of artificial intelligence in the Olympic movement. Developed in collaboration with digital transformation partner Deloitte, the initiative sets out seven core principles, including data protection, accountability, fairness and sustainability.
The system also includes a practical checklist designed to help sports organizations assess and mitigate the risks associated with AI deployment. Alejandro Moreno Madrid, head of the IOC's artificial intelligence department, said that while the technology has the potential to transform athlete training and fan engagement, it must remain subordinate to the needs of the sport.
He stressed that the system serves as a common framework to ensure that athlete data and competition integrity remain central as its implementation grows. The IOC said in a statement that the leadership is responding to growing concerns about security and transparency in various areas, from game operations to daily organizational workflows.
The move is in line with a broader industry trend where regulators are setting ethical barriers to manage the rapid integration of underlying models and automated systems into high-stakes sectors. News #14. Microsoft has launched a new set of data tools designed to help businesses integrate their own operational knowledge into AI-powered workflows .
The company positions these innovations to go beyond the capabilities of one-size-fits-all models, focusing on the specific institutional data and decision-making models that define the business. By prioritizing this proprietary context, Microsoft seeks to differentiate its enterprise offerings in a market where basic models are increasingly commercialized.
While competitors like Google and Amazon continue to emphasize universal artificial intelligence, this move signals a strategic shift toward deep business integration. The company argues that the true value to the enterprise lies in teaching systems the unique operational experience that distinguishes one organization from another.
This is a clear step towards reaching the next stage of enterprise AI implementation, where competitive advantage lies in how well a firm uses its own internal data. News #15. President Donald Trump is hosting a summit with top AI executives in Washington on Tuesday to discuss the future of the industry.
The meeting is attended by prominent leaders such as Mark Zuckerberg of Meta, Dario Amodei of Anthropic, and Greg Brockman of OpenAI, as well as Speaker of the House Mike Johnson. While these leaders have recently advocated for increased government oversight and global regulation to mitigate potential dangers , the president has publicly dismissed these existential concerns as a hoax.
Instead, he insists that American firms support rapid innovation to ensure the US has an edge over China in the global artificial intelligence race. During the same visit, the president will also launch america.gov, a new portal with an artificial intelligence-based tool designed to help citizens access government services.
This summit highlights the growing tension between industry leaders who call for protective fences and an administration that prioritizes aggressive development . The event comes as voters weigh the impact of these technologies ahead of the midterm elections. News #16: Abu Dhabi's sovereign wealth fund Mubadala Investment Company is partnering with American AI to develop artificial intelligence infrastructure in the United Arab Emirates.
The collaboration focuses on identifying opportunities to support the deployment of AI across the country, which aligns with the UAE’s broader strategy to build sovereign capacity and establish itself as a global hub for technological innovation. As part of this agreement, Together AI will provide a physical presence in Abu Dhabi.
The move aims to directly transfer the firm's specific capabilities in inference and research to developers building AI applications across the region. Ali Al Muhairi, Executive Director of Diversified Assets at Mubadala UAE Investments, said the partnership is designed to deepen local expertise and translate innovation from research into real-world applications .
By integrating Together AI's infrastructure into the local ecosystem, the agreement aims to expand technical access for regional developers. This initiative is the result of the UAE's ongoing efforts to leverage its diversified energy balance and international partnerships to accelerate the domestic AI value chain .
News #17: Microsoft and the Philippine Department of Education are expanding their partnership to integrate artificial intelligence into the nation's public school system. As part of this initiative, Microsoft is providing access to Copilot to 1 million teachers in Philippine public schools. The deployment aims to assist teachers with lesson planning, administrative tasks, and personalization of instructional materials to enhance classroom learning.
Education Secretary Sara Duterte said the tools are designed to support the resumption of learning and prepare students for an AI-driven economy . This collaboration builds on a long-standing relationship between the two organizations and is in line with the Ministry of Education's five-point program to modernize the country's academic infrastructure.
By reducing the administrative burden on staff, the program intends to increase opportunities for direct student engagement. The move is the result of years of prior collaboration between the government and the tech giant on digitizing Philippine classrooms . News No. 18. The United Nations Security Council officially recognized artificial intelligence as a key international security issue during a special session held on September 23.
This meeting marks a shift from previous debates that focused largely on labor market disruptions and the impact on education, to integrating AI into global strategic risks such as war, cyberattacks, and deterrence. The Council's decision to include this technology on its agenda signals that AI is now seen as a major factor in maintaining international peace.
The session went beyond traditional diplomatic discourse to include machine learning researcher Joshua Benchio and executives from leading AI development firms. By bringing these technical experts directly into the room, the UN is treating the development of basic models and automated systems as a matter of global competition between states.
This transition highlights that policymakers no longer view AI as simply a productivity tool, but as a critical element of national and international security infrastructure. As nations address these strategic risks, the conversation has officially moved from the inner labs of the tech industry to the highest levels of global governance.
Thanks for watching today's AI news. If you enjoyed this update, please click the "Like" button and subscribe to stay one step ahead. Tomorrow we'll dive into the latest breakthroughs in generative video technology that you won't want to miss. See you in the next issue.
Comments 0