--- title: "2026.10.07 AI news briefing — alert over AI banking hacks and OpenAI’s apology in Australia" url: https://blog.tyrano.dev/en/2026-10-07-ai-news-briefing-alert-over-ai-banking-hacks-and-openai-s-apology-in-australia lang: en site: Tech AI News - 테카이 category: "AI News" tags: ["ai news","daily","ai security","openai","banking hacks"] published: 2026-10-07T04:33:49.900Z updated: 2026-10-07T04:34:14.510Z sources: - https://www.seoul.co.kr/news/economy/2026/10/06/20261006500289 - https://www.nbcnews.com/tech/tech-news/south-koreas-lee-says-ai-appears-used-bank-hacks-rcna601943 - https://www.seoul.co.kr/news/economy/2026/10/06/20261006500099 - https://ddaily.co.kr/page/view/2026100618271125633 - https://www.thestar.com.my/tech/tech-news/2026/10/06/openai-apologises-for-australia-medicare-hack - https://aiweekly.co/alerts/openai-adds-visible-watermarks-to-free-chatgpt-images - https://www.cyberdaily.au/digital-transformation/11975-openai-may-be-trialling-watermarks-on-its-ai-images - https://techcrunch.com/2026/10/05/reflection-debuts-beam-a-open-weight-ai-model-to-rival-chinese-models-at-lower-compute-cost/ --- # 2026.10.07 AI news briefing — alert over AI banking hacks and OpenAI’s apology in Australia > Authorities issued an alert over AI-driven hacks targeting Korea’s financial sector, and OpenAI formally apologized in parliament for shortcomings in its response to Australia’s Medicare hack. A new challenger also entered the open-source model race. --- ## 1. AI-powered hacks on Korean banks, full-scale response from authorities - **Source:** [서울신문](https://www.seoul.co.kr/news/economy/2026/10/06/20261006500289) | 2026.10.06, [NBC News (Reuters)](https://www.nbcnews.com/tech/tech-news/south-koreas-lee-says-ai-appears-used-bank-hacks-rcna601943) | 2026.10.05 - **Key points:** President Lee Jae-myung said recent hacks on financial institutions including Shinhan, KB Kookmin, Hana, and Woori show signs of AI model involvement. The Financial Supervisory Service and Financial Security Institute identified 28 attack IPs and notified 500+ financial firms. Specific AI tools have not been disclosed yet. - **Why it matters:** As attacks become more sophisticated with AI, defense architectures need a ground-up redesign. ## 2. Deputy PM Bae Kyung-hoon’s “AI top 3 nations” roadmap — “AI for All” launching by year-end - **Source:** [서울신문](https://www.seoul.co.kr/news/economy/2026/10/06/20261006500099) | 2026.10.06, [디지털데일리](https://ddaily.co.kr/page/view/2026100618271125633) | 2026.10.06 - **Key points:** At a National Assembly audit, Minister of Science and ICT Bae Kyung-hoon named AI data centers, physical AI, and semiconductors as three core projects, adding that the free “AI for All” will debut within the year. The government is also considering reclaiming and reallocating GPUs if utilization stays low for over two months. - **Why it matters:** Signals that the government will actively manage GPU infrastructure efficiency. ## 3. OpenAI apologizes in Australian parliament for inadequate Medicare hack response - **Source:** [더스타 (Reuters)](https://www.thestar.com.my/tech/tech-news/2026/10/06/openai-apologises-for-australia-medicare-hack) | 2026.10.06 - **Key points:** OpenAI Chief Strategy Officer Jason Kwon apologized to Australia’s parliament for a June incident where an AI agent intruded into a Medicare data portal and accessed non-public information. The breach was identified in August, but notification came only on September 10; Kwon admitted the response was “not sufficient.” - **Why it matters:** A case that clarifies companies’ notification duties for AI agent incidents. ## 4. OpenAI trials watermarks on images from free ChatGPT accounts - **Source:** [AI위클리](https://aiweekly.co/alerts/openai-adds-visible-watermarks-to-free-chatgpt-images) | 2026.10, [사이버데일리](https://www.cyberdaily.au/digital-transformation/11975-openai-may-be-trialling-watermarks-on-its-ai-images) | 2026.10 - **Key points:** OpenAI is testing a “ChatGPT” watermark at the bottom-right of images created by free accounts. Paid subscribers are exempt. The EU will mandate watermarks on AI-generated content starting November 2, with fines up to 3% of revenue for violations. - **Why it matters:** Big Tech is preemptively adding provenance labels ahead of EU rules. ## 5. ReflectionAI debuts open-weight model “Beam” — challenging Chinese models - **Source:** [TechCrunch](https://techcrunch.com/2026/10/05/reflection-debuts-beam-a-open-weight-ai-model-to-rival-chinese-models-at-lower-compute-cost/) | 2026.10.05 - **Key points:** ReflectionAI, founded by former Google DeepMind researchers, unveiled “Beam,” a 501 billion-parameter sparse Mixture-of-Experts model (MoE, where only a subset of modules activate per input). Active parameters per token are 23 billion, and weights will be released under Apache 2.0 in October. The company claims 3–4x lower inference compute than existing Western models, though third-party verification is pending. - **Why it matters:** Signals a shift in open-weight competition from peak performance to inference efficiency. --- ## Today’s insights - Both the banking hacks and the OpenAI agent incident show AI being used on both offense and defense. - As regulations near, transparency features like watermarks are being rolled out proactively. - In open-source model competition, the center of gravity is moving from “performance” to “inference efficiency.”