[DigitalToday reporter Chi-gyu Hwang (황치규)] The debate over the stability of AI models has flared again. As people who work at, or previously worked at, AI companies talk about the possibility that AI could wipe out humanity, the AI safety narrative is unfolding on a different level than in the past. Major AI companies have even raised the idea of slowing AI development. Sam Altman (샘 알트먼), CEO of OpenAI, said addressing AI safety issues comes first, and said it will not pursue an initial public offering (IPO) this year.

ㆍOpenAI also reviews slowdown…U.S. AI industry sees safety worries spread after security incidents ㆍSam Altman: "No OpenAI listing this year"…AI safety comes first ㆍ[Tech Insight] "If we build AI models the way we do now, we cannot solve AI safety issues"

Some interpret the AI-driven human extinction argument as fear marketing by Anthropic, which is seeking a $2 trillion valuation and is preparing for an IPO. Jensen Huang (젠슨 황), CEO of Nvidia, has said risk arguments around AI are a tool to create new demand for cybersecurity. At an event, he said, "The reason there is so much talk about cyber security these days is that the industry is preparing to launch new products," and added, "What better way is there to create demand than to create a problem?"

ㆍWhistleblowing or fear marketing?...Sudden debate over the probability of AI-driven human extinction ㆍNvidia CEO dismisses AI risk arguments..."Fear marketing to sell security products"

OpenAI's new GPT-6 Astra made an impressive debut in the AI field. Praise has poured in from people who have tried Astra.

ㆍGPT-6 Astra clears 3D game 'Portal' without human help…cost 760,000 won ㆍGPT-6 Astra ranks near the top with a 67 coding score…stands out for cost efficiency per task ㆍJensen Huang: "Congratulations to OpenAI on Astra…AGI has arrived, next is 400,000 GPUs" ㆍOpenAI's GPT-6 Astra dominated in 3D and games but suffered a 'humiliation' in writing

OpenAI also temporarily halted new sign-ups for its $200-a-month Pro plan due to a surge in demand for Astra. There is also an assessment that while performance improved with Astra, it has become more difficult to monitor how it works.

ㆍOpenAI temporarily halts Pro subscription sign-ups amid surge in Astra demand ㆍ"OpenAI's Astra has improved performance, but is harder to monitor" ㆍHas AI started to 'think to itself'?…Opaque recursion spotlighted by Astra

How Astra could change the way AI is used has also emerged as a point to watch.

ㆍ[Tech Insight] In the GPT-6 Astra era, rethink skills and prompts

It also summarised moves by domestic and overseas companies around AI.

Samsung SDS unveiled its enterprise AI strategy based on a "Multi-Dimensional AI full-stack" strategy at its annual AX conference, REAL Summit 2026.

ㆍSamsung SDS bets on tailored AX based on a multi-dimensional AI full stack...expands FDE to about 200 people ㆍSamsung SDS steps up physical AI push...to deploy a robot orchestration platform next year ㆍSamsung SDS unveils AI agent service 'Brity Works Co-work'

LG AI Research and the National Pension Service signed a memorandum of understanding (MOU) to jointly develop AI to improve the efficiency of pension fund management. Twelve Labs provides its video embedding and search model Marengo via an Amazon Bedrock managed knowledge base. MegazoneSoft completed building an AX system based on Google Workspace for cosmetics ODM company C&C International. GazonCloud will build an AI agent for automating quality assurance (QA) work at Hanmi Pharmaceutical, called Hanmi Q-Agent, based on its enterprise AI OS, AIR Studio. Animorp, a full-stack GEO platform company, unveiled an API for generative AI search optimisation (GEO), querying.ai. Younglimwon Soft Lab is accelerating efforts to link employee-led AI use experiences to advancing its ERP products.

ㆍLG AI Research to cooperate with National Pension Service on fund-management-focused AI development ㆍTwelve Labs video model Marengo supports Amazon Bedrock knowledge base ㆍMegazoneSoft builds companywide AX system for C&C International based on Google Workspace ㆍMegazoneCloud builds Hanmi Pharmaceutical QA AI agent based on AIR Studio ㆍAnimorp unveils GEO API querying.ai...provides actual answers and sources on AI search screens ㆍYounglimwon Soft Lab pushes employee-led AI-driven ERP advancement in earnest

OpenAI unveiled ChatGPT for Financial Services, tailored to investment banking work. OpenAI also released ChatGPT Image 2.5, which improved ChatGPT's image generation capabilities.

ㆍOpenAI targets Wall Street junior bankers with Astra-based ChatGPT for Financial Services ㆍChatGPT images have changed markedly…OpenAI unveils Image 2.5 with up to 50 percent faster generation speed

China's AI company Moonshot AI has reportedly set a goal of reaching an annualised revenue run rate of $2 billion by the end of the year. That is twice the Moonshot AI revenue run rate known as of August.

ㆍKimi developer Moonshot AI targets $2 billion in annual revenue

SAP CEO Christian Klein (크리스티안 클라인) said it would focus only on developing AI agents that automate the most complex enterprise business processes to differentiate itself in the market. Salesforce unveiled an enterprise AI harness and an AI control plane that support building and managing AI agents. Lightfield, which bills itself as a CRM for AI agents, raised $47 million in Series A funding. Nvidia will work with Palantir to introduce sovereign AI into supply chains, starting with Nvidia's own operations. Red Hat unveiled a new version of its AI platform, Red Hat AI 3.5. Red Hat AI 3.5 strengthened pre-deployment security verification, multitenancy for shared GPU infrastructure, observability and agent development functions.

ㆍSAP CEO: "We will only release AI agents that solve complex problems" ㆍSalesforce unveils enterprise AI harness...also announces AI monitoring service ㆍLightfield, a CRM for AI agents, raises $47 million ㆍNvidia to work with Palantir...introduces sovereign AI into core supply chains ㆍRed Hat unveils enterprise AI platform Red Hat AI 3.5...strengthens security verification and GPU multitenancy

Amazon officially launched Quik, an enterprise AI assistant and AI agent platform, for macOS and Windows. Meta launched Muse, a personal AI agent app internally known by the codename Hatch, for U.S. consumers.

ㆍAmazon officially launches desktop version of enterprise AI agent Quik...integrates mail, calendar and CRM ㆍMeta launches personal AI agent Muse…also introduces $20 and $100 monthly plans