Afternoon edition — Sunday 1 November 2026

A developer documentation page from xAI, whose docs now carry the SpaceXSI name, says that starting 7 November 2026 text responses from grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants will include an invisible watermark. The page says the watermark is applied "to comply with the EU AI Act". It was last updated on 31 October.

The watermark is on by default on every API endpoint. It covers ordinary text, structured outputs and tool-call arguments, and a response header, x-watermark, shows whether it was applied. Team admins can opt out per model, but must confirm that a legal exception applies and that they are responsible for any marking or disclosure duties.

The mark works by slightly shaping which tokens the model picks, so no characters are added. Detection is statistical and returns watermarked, not watermarked or uncertain, with longer text giving a stronger signal. Access to the detection API is granted by application.

An OpenAI Help Center page says that starting in December, the usage allowance on Plus, Pro and Business plans will be shared across ChatGPT and Codex, "as we bring the agentic capabilities of Work into the main ChatGPT experience". Usage will be measured in tokens, so more advanced models and complex tasks use more of the allowance; Instant chat will not draw from it.

OpenAI says it expects most users will not have to wait for a reset more often than they do now. Combined remaining usage and reset times will appear under Settings, and users who hit the limit can wait, upgrade or buy credits. The page gives no allowance figures.

BleepingComputer, which saw an OpenAI email to subscribers, reports that a combined Chat and Work experience will be tested with a small group of users in the coming weeks, and that Codex stays a separate experience in the ChatGPT desktop app. Hacker News users were discussing the help page by late Saturday evening, after this journal's Saturday afternoon edition.

A second help page sets card and device limits: too many Go, Plus or Pro subscriptions on one payment card or device can block a purchase, and a device can show "Device limit reached". Free, Business, Enterprise and Edu accounts do not count. The Decoder reads this as aimed at people who rotate several cheap paid accounts to get more usage.

The Observer reports that Britain's most senior mathematicians, including three Fields medal winners, met at the Royal Society in London last week. The meeting followed OpenAI's publication last month of solutions, generated by its frontier models, to 377 of the hardest mathematical problems.

Martin Bridson, president of the Clay Mathematics Institute, described a "longer-term fear that knowledge could be hoarded behind paywalls" and asked: "Why should they get to say who has access to these tools, and on what basis do they choose who gets access?" So far OpenAI and Anthropic have each committed $5m to a Simons Foundation fund that gives some researchers access to their models.

Fields medallist Martin Hairer said open models cannot compete with frontier ones, leaving "one of the main tools for mathematic research" under the control of two US companies. Ana Caraiani of Imperial College London said no institution in Romania, her native country, is eligible to apply for access.

Axios reports that right after Tuesday's US midterm elections, OpenAI will launch a policy and PR campaign starting in Congress's lame-duck session, which opens on 9 November. Chris Lehane, OpenAI's chief global affairs officer, told Axios his team has been stockpiling proposals since August, "some new and some repackaged".

The first proposal is what OpenAI calls "speed limits": new ideas for mandatory safeguards on the most powerful AI systems, with a more aggressive drive in Congress. "You're going to see us talk more and more about the concept of speed limits," Lehane said.

On data centres, OpenAI plans community compacts and proposed state legislation giving residents a say in how projects are built and what benefits they receive. It will also argue that its facilities should be called "AI compute campuses", citing a project in Pike County, Ohio. Further proposals cover help for workers, education and training, and US-shaped international standards.

KB Kookmin Bank said on Sunday that it holds the mandate to lead a 10.7 trillion won ($8 billion) syndicated loan for a Brookfield-backed AI computing project, The Korea Herald reports. The bank calls it South Korea's first GPU financing and the country's largest-ever syndicated loan arrangement.

The loan is part of a project worth about 13 trillion won. Brookfield will set up a special-purpose company, tentatively named Asset Co., to buy GPUs and other computing assets and supply capacity to Naver Cloud under a long-term agreement. The bank named no other lenders and gave no loan terms.

Naver's filing sets tentative targets of 55 megawatts in the first half of 2027, 100 megawatts by year-end and 200 megawatts in 2028, centred on its GAK Sejong data centre. Naver signed its $9 billion infrastructure agreement with Brookfield on Tuesday, a condition of Nvidia's planned $1 billion investment in Naver.

Korea's Financial News reports that Korea Investment Holdings and Japan's SBI Group plan a joint fund to invest about 1.15 trillion yen (about 9.75 trillion won) in Japanese data centres with 200 megawatts of capacity. The two groups are sounding out companies in both countries as investors and plan a blueprint by the end of the year.

Companies are looking at supplying chips, cooling and power equipment and at building and operating the sites. Kyushu is one candidate location, and the project's backers are understood to have held talks with Kyushu Electric Power.

Maeil Business reports a fund target of more than 1 trillion yen and quotes Korea Investment as saying the plan "has not been confirmed". KED Global says the groups hope to expand the partnership over 30 years.