Anthropic's background agent 'Claude Duet', the general availability of OpenAI's Decisions API, the move of agent standards to a neutral foundation, and a U.S. federal order to remove DeepSeek — a roundup of this week's key AI industry stories.
Anthropic has opened a waitlist for 'Claude Duet', a proactive agent app that keeps working while the user is away. Built on Claude Opus 5.5, it starts as a preview for U.S. Pro and Max subscribers. The move counters OpenAI and Meta in the race for personal agents, showing that agents are evolving from 'assistants that wait' into 'colleagues that act first'.
OpenAI's Decisions API has entered general availability (GA). The GPT-6 Luna-based API returns bounded answers such as probabilities, choices, and scores, costing $0.10 per 1 million input tokens with no output-token charge. A single request can handle up to 400 questions, so for fixed-answer decisions — like branching judgments in an agent — it offers a fast, cheap alternative to generating long sentences. Microsoft has also begun offering the Azure OpenAI Decisions API as generally available in Foundry.
At AGNTCon + MCPCon in San Francisco on October 22, the Linux Foundation's Agentic AI Foundation (AAIF) announced it will take on four new agent standards. The stars are Anthropic's Agent Skills, OpenAI's Open Responses, Hugging Face's OpenEnv, and Alibaba's OpenSandbox. Following MCP and Google's Agent2Agent protocol, core agent specifications are now gathering under a neutral body rather than any one company, opening the way for developers to write a skill once and reuse it across multiple models and platforms.
This was the week when 'where you buy frontier models' changed. OpenAI's frontier models and Codex came to Google Cloud's Model Garden, with GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna listed in private preview. Anthropic also made its full Claude Platform generally available on Google Cloud, so customers can pay through existing cloud commitments. Microsoft remains OpenAI's primary cloud partner — that hasn't changed.
The U.S. Department of Homeland Security (DHS) published an order in the Federal Register on October 20 requiring federal civilian agencies to remove all DeepSeek products from their IT systems. The order covers not only DeepSeek's own products but also third-party products and services that use DeepSeek technology, banning new purchases and requiring removal from existing systems. DHS warned that DeepSeek models carry 'multiple threat vectors' and that sensitive data could be transferred to China. The order applies only to civilian agencies, however, and does not directly regulate private companies or ordinary users.
Korean AI news kept coming. Samsung Electro-Mechanics won a multi-billion-dollar contract to supply FC-BGA substrates for Nvidia's AI chips, and LG Group hired a robotics chief from Hyundai Motor Group to strengthen its physical AI strategy. Naver invested in Instinct, a U.S. AI assistant startup valued at $10 billion, and music generator Suno, stepping up its global expansion. Meanwhile, the KOSPI fell as concerns about AI spending grew amid Meta's worsening cash flow, and AI cloud company Elice plunged more than 30% on its first day on the KOSDAQ, spreading the debate over AI valuations to the Korean market as well.