[10/28 09:00 UTC] Anthropic says an unreleased internal model found flaws in Chrome, Safari,…

AI News Flash for October 28, 2026 at 09:00 UTC. Topics: 1. Anthropic says an unreleased internal model found flaws in Chrome, Safari, Android, Windows and Linux and chained them into working exploits https://thehackernews.com/2026/10/anthropics-unreleased-ai-model-chains.html 2. Trend Micro details three cyberattacks on Japanese organizations where AI played a key role, including a campaign run by 44 AI agents https://thehackernews.com/2026/10/cyberstrikeai-uses-44-ai-agents-to.html 3. China is expected to adopt its AI export curbs by year-end, as top labs seek clarity on training models on overseas servers https://www.reuters.com/business/retail-consumer/chinas-new-export-curbs-draw-line-around-its-ai-crown-jewels-2026-10-28/ 4. Anthropic study finds professional analysts went along with their AI agent's hidden lean 86 percent of the time https://www.aljazeera.com/economy/2026/10/28/experts-defer-to-ai-in-decision-making-anthropic-study-finds 5. Japan launches a working group to consider revising its AI law, including penalties for companies that ignore government requests https://sp.m.jiji.com/article/show/3899202