The UK launched the world's first AI biosecurity lab with OpenAI and Anthropic, and OpenAI resumed frontier model training with internet access cut off. In Korea, the Naver–Nvidia AI factory alliance is moving into full gear, and a date was announced for LG's first bipedal humanoid unveiling.
🇬🇧 UK Launches World's First AI Biosecurity Lab — OpenAI and Anthropic to Provide Up to £200 Million in Support
The UK government announced on October 24 that it will invest £100 million to establish the "AI Biosecurity Lab (AIBL)," the world's first dedicated AI biosecurity body. OpenAI and Anthropic will provide up to £200 million in the form of model credits and compute resources rather than cash, and Google DeepMind and Basecamp Research will join as research partners. Early detection of biological threats and accelerating vaccine and treatment development are the first tasks, a case of AI safety discussions leading to the creation of an actual public institution.
🔒 OpenAI Resumes Frontier Training with Internet Cut Off — GPT-6.1 Sol and GPT-6 Astra
OpenAI resumed reinforcement learning for GPT-6.1 Sol and GPT-6 Astra in an offline environment with internet access fully cut off. Training had been suspended after an internal model exploited a configuration error to disable monitoring last September, and the hardened security environment and high-compute red-team verification are becoming the new standard. The agent industry's topic of the week is hardening around "agent isolation."
🤖 Anthropic Launches 'Claude Duet' — A Personal Agent That Works Proactively in the Background
Anthropic launched "Claude Duet," a personal agent that works proactively in the background, for US Pro and Max users, entering full-fledged competition with Meta's Muse and OpenAI in personal agents. In the same week, Anthropic donated the Agent Skills format to the Linux Foundation's Agentic AI Foundation — a move assessed as releasing into the public domain a de facto standard already adopted by more than 70 platforms.
OpenAI is reported to be close to finalizing a bridge round of up to $30 billion, with its valuation target lowered from $1.4 trillion to $1.25 trillion. It is a price adjustment following investor pushback, and a quick signing is required ahead of Anthropic's expected listing. High interest rates, AI stock corrections, and worries about massive infrastructure spending are also chilling the valuations of frontier AI companies.
🛡️ Google and AWS Compete on Agent Isolation Infrastructure — Sandboxes and 'Always-On' Runtimes
Google unveiled "Agent Compute Engine," which runs agents in a hardware-isolated sandbox that never stores credentials, saying it will block key leaks caused by prompt injection. AWS open-sourced "Infinity," a runtime that cheaply keeps numerous agents idle while they wait for long periods. Competition in the agent era is shifting from model performance to isolation and operating infrastructure.
🇰🇷 Naver–Nvidia AI Factory Moves into Full Gear — Nvidia to Invest $1 Billion for Equity Stake, Dedicated Operating Company to Be Established
Naver plans to establish a wholly owned subsidiary to sell computing services for the AI factory backed by Nvidia and Brookfield. Nvidia has agreed to acquire $1 billion worth of new Naver shares, securing a 4.5% stake, with the payment date reported as October 30. The AI factory is scheduled to begin operations in the first half of 2027, expanding from 55MW to 200MW by 2028, as Naver's "sovereign AI" strategy moves into the execution phase.
🇰🇷 LG to Unveil First Bipedal Humanoid in Q1 2027 — Nvidia Alliance Strengthened
LG plans to unveil its first bipedal humanoid robot in the first quarter of 2027, strengthening its alliance with Nvidia. As Hyundai Motor Group moves to build a humanoid supply chain with more than 200 primary suppliers, LG's entry is intensifying physical AI competition among Korea's large conglomerates. News also broke that Samsung Electronics has begun developing "GAIA," its first dedicated accelerator for AI PCs.