Today's Headlines

  • US Homeland Security publishes an order removing DeepSeek products from federal civilian agencies, including third-party products built with DeepSeek
  • Japan's Digital Agency launches AI agent tools for about 180,000 civil servants on its Gennai platform and publishes the rules that govern them
  • Shared conversations from Meta's Muse could be found through DNS records, and Meta pauses sharing
  • OpenAI acquires Orchestra Bio, a platform for coordinating drug development

The day's thread is where lines get drawn around AI. The US Department of Homeland Security published an order that removes DeepSeek's products, and third-party products built with its technology, from federal civilian agencies. In Japan, the Digital Agency began giving some 180,000 government employees tools to build and run their own AI agents and, on the same day, published rules setting out who is responsible for what.

Today's Top Three

US Homeland Security orders DeepSeek products out of federal civilian agencies, including third-party products built with DeepSeek

The US Department of Homeland Security (DHS) published a notice in the Federal Register on October 20 of an order removing products from China's DeepSeek from the procurements and IT systems of federal civilian agencies.

The order reaches both procurement and systems already in place. On procurement, IT products and services developed by DeepSeek and its affiliates are excluded from civilian agency purchasing.

On systems, civilian agencies must remove products from DeepSeek and its affiliates, as well as third-party products and services developed using DeepSeek implementations. Because the order reaches third-party products, products and services with DeepSeek models built in also come out of federal civilian IT systems.

The order rests on the Federal Acquisition Supply Chain Security Act of 2018. The signed order attached to the notice carries the signature of Homeland Security Secretary Markwayne Mullin and an original release date of July 24. It applies to federal civilian executive branch agencies and leaves out their national security systems.

The order cites DeepSeek's servers being located in China and a Chinese legal system that gives the government broad access to private company infrastructure. It also points to the collection of personally identifiable information at account sign-up, the potential to track users, and outside research finding multiple critical vulnerabilities in DeepSeek's iOS app. Agencies can request an exception in writing, and CISA reviews the requests.

Japan's Digital Agency launches AI agent tools for about 180,000 civil servants and publishes the rules that govern them

Japan's Digital Agency began rolling out two environments for building and running AI agents on October 20, on Gennai, the generative AI platform it runs for government employees.

The rollout covers about 180,000 civil servants. Agent Chat, which lets staff build and run simple agents, started reaching ministries and agencies that day. Gennai Kobo, a workshop for building more advanced agents, is due to arrive by around December.

Gennai Kobo works like an AI coding agent, taking natural-language instructions and carrying out code generation and edits, file operations, command execution and testing on its own. The agency says its own research shows this is the world's first rollout of tools at this level to ordinary staff, not just specialists, as shared government infrastructure for some 180,000 people.

The agency also published the Gennai standard on governance, security and safety, the rules for using the new tools. It sets out responsibilities and requirements separately for users, administrators and developers, and was drawn up with help from Japan's AI Safety Institute (AISI).

The Digital Agency says the standard can serve as a model for local governments and private companies weighing AI agents of their own. Both environments are due to be released around February 2027 as Gennai OSS Ver. 2.0, open-source software that anyone can use, modify and redistribute.

Shared conversations from Meta's Muse could be found through DNS records, and Meta pauses sharing

Security firm Claroc reported that conversations users shared from Meta's AI agent Muse could be found by people who were never given the link.

When a user shares a conversation, Muse publishes a page whose hostname combines the conversation's title with a random ID. The first time anyone opens that page, the hostname is logged as a DNS lookup and becomes searchable in commercial passive-DNS services.

Searching for the domain that serves the shares, Claroc found more than 6,800 share hostnames. It opened a few hundred, most of them pages their owners plainly meant to be public. A minority were not, including a candidate's notes for a confidential executive-search call with a frontier AI lab, an internal revenue review marked for a company's CEO, a small-business loan application and a workplace's full-year staff vacation calendar.

Claroc reported the issue through Meta's bug bounty program on October 8. After Ars Technica asked about it, Meta said on October 20 that it had paused the sharing feature while it reviews the report and considers changes to how the links can be discovered.

Claroc advises companies to treat an AI share link as an outbound transfer of whatever was in the conversation. Where they have no way to govern it, the firm suggests blocking muse.ai and its content domain at the proxy or DNS filter.

Muse launched in September. According to TechCrunch, Sensor Tower data show 10.9 million downloads since launch, 10.2 million of them in the US.

Other Developments

Models

Research

  • Anthropic published research on October 20 in which an internal research version of Claude designed proteins that bind scorpion neurotoxins. Over 64 hours, Claude generated 318,484 designs; of the 211 designs lab-tested by two outside firms, 17 met Anthropic's criteria. All 211 designs and their data have been released. How Claude designed protein binders against scorpion neurotoxins (Anthropic)

Products

  • OpenAI added the option on October 20 to lock ChatGPT conversations with a four-digit PIN, Face ID or Touch ID. Locked chats move to their own section, apart from regular history and chat search. ChatGPT now lets you lock your chats (TechCrunch)

Policy

  • Meta removed the word "train" and the phrase "any content" from the new commercial terms that take effect on October 30. The revised wording lets Meta periodically pull updates from business users' websites to provide, develop, improve and research advertising and business products. According to Ars Technica, legal experts say "develop" remains broad. Meta drops the word "train" from its commercial TOS after AI backlash (Ars Technica)
  • In copyright suits brought by book authors and a publisher, Magistrate Judge Thomas Hixon ordered Meta to identify the people behind decisions to download books from shadow libraries. According to Ars Technica, Meta had argued that staff downloaded the torrents themselves and that nobody with authority made a decision. Meta tries to scapegoat staff who downloaded torrents to spare execs (Ars Technica)

Business

Source: Selected by the editors from the AI news inbox (collected October 21, 2026: 56 items, 10 primary and 46 secondary).