AI News Nov 2, 2026

On November 2, 2026, Abu Dhabi's Media Office reported that Sheikh Khaled bin Mohamed bin Zayed Al Nahyan met Sam Altman and witnessed four agreements with OpenAI. The energy agreement deploys GPT models at ADNOC, with inference kept inside the UAE. OpenAI's help center says Plus, Pro, and Business plans will share one token allowance across ChatGPT and Codex starting in December. Google DeepMind is opening a 12-month PlanAssist beta that adds twelve English councils to the three already in the pilot. On November 7, SpaceXSI will watermark text from supported Grok API models by default.

In the news

Four agreements in Abu Dhabi

The Media Office account of the November 2 meeting covers government, energy, and investment, and it quotes Sam Altman. That release is the record of the signing. OpenAI's news index has no November 1 or November 2 item for the meeting. The release names no price, no GPU count, and no model version.¹²

The energy agreement brings together ADNOC, Inception42 (a subsidiary of G42), and OpenAI. The Media Office calls ADNOC the first company in the energy sector to deploy OpenAI's frontier technology, and it says the work will feed operational, investment, and commercial decisions. Deployment is on OpenAI's GPT models, with inference kept inside the UAE. The release does not name which GPT model. It does not say that this signing makes ChatGPT a default tool across the UAE.¹

The government agreement launches what the release calls the world's first AI-native Machine-Readable Government initiative. A defined set of policies and regulations is translated into machine-readable form, so AI systems can help officials with routine compliance, including permits and certificates. Each assessment is meant to trace back to the governing rule, and officials keep oversight and the final decision. The Department of Government Enablement, Abu Dhabi, leads the work, and the first pilot is already in production at the Abu Dhabi Agriculture and Food Safety Authority on food imports, using GPUs in UAE data centers under OpenAI's in-country inference residency. The release names no other production workload.¹

Mubadala, Inception42, and OpenAI signed an agreement to use AI for investment analysis, identifying opportunities, and assessing risk. The release states no assets under management and no performance result. Inception42 and OpenAI also signed a five-year deal covering government, enterprise, and education in the UAE, including enterprise tools for large organizations there. That five-year deal is separate from the earlier Stargate UAE partnership.¹

One allowance for ChatGPT and Codex

OpenAI's help article, Understanding ChatGPT and Codex usage limits, says that starting in December the usage allowance on Plus, Pro, and Business plans will be shared across ChatGPT and Codex. OpenAI ties the change to bringing the agentic capabilities of Work into the main ChatGPT experience. Usage is measured in tokens, so a more capable model or a more complex task spends more of the allowance. Instant chat does not draw from it. The article sits in the help center, and OpenAI's news index does not carry it.³²

People will see the combined remainder and the reset time under Settings, then Usage. At the limit, the choices named on the page are to wait, to upgrade, or to buy credits where a feature allows it. OpenAI says it expects most people will not need to wait for a reset more often than they do today. The article gives no token budget, no multiplier, and no calendar day in December, and it does not extend the change to Enterprise, Edu, or Free.³

Work and Codex already shared one agentic allowance, separate from ordinary chat. In December, ChatGPT conversations other than Instant join that same allowance, and Codex stays on it. The article does not say the Codex desktop view is being removed, and it does not announce a new price.³

PlanAssist moves to a wider beta

On October 31, 2026, Google DeepMind said PlanAssist is leaving a three-council pilot for a 12-month beta on live applications. Owen Larter and David Thacker wrote the post on the Responsibility and Safety team. The tool, built with Gemini and with the UK government, Google Cloud, and Faculty, drafts paperwork for routine householder planning applications. A qualified officer still reviews and approves each assessment before any decision. The post does not announce a new Gemini model.⁴

In Barnet, Camden, and Dorset, average officer time on householder assessment tasks fell by 58 percent. Officers accepted more than 70 percent of the tool's reports with no edits. The reports cover whether an application follows local rules and whether the required documents were included. Householder applications make up as much as 70 percent of planning officers' workloads, which is the scale the time cut applies to.⁴

The beta will bring in twelve more councils, in three cohorts, for 15 councils in total, mixing city and rural districts. The post does not name the twelve. If the beta goes well, the government plans to offer the tool to every council in England in 2027. Before a report is drafted, the system checks documents and drawings, heritage limits including tree preservation orders, local and national policy, and earlier objections. The officer can still edit the cited report that comes out.⁴

A June 2026 DeepMind post announced the prototype in the same three councils. The October 31 post is the scale-up, with the 58 percent and 70 percent figures and the year-long beta. The post frames faster planning decisions around the UK government's aim of 1.5 million homes by 2029. That aim is the page's framing, not a commitment that DeepMind will deliver those homes.⁴

Watermarks on Grok API text

SpaceXSI's developer docs say that on November 7, 2026, text responses from supported Grok models will carry an invisible watermark by default, so the company can later recognize text a model produced. The reason given on the page is the EU AI Act. Callers do not pass a parameter to turn it on. The response body keeps its shape, and an x-watermark header reads true when that response was marked. The docs index dates the page October 31, 2026, and the page itself gives no clock time.⁵

The models listed are grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20-reasoning, grok-4.20-non-reasoning, and grok-4.20-multi-agent. The watermark applies on every text generation API, including Responses and Chat Completions, on https://api.x.ai and on regional endpoints. It covers ordinary text, structured outputs, and the arguments passed to tools. SpaceXSI says this does not change quality or speed.⁵

Team administrators can turn the watermark off in the SpaceXSI Console, one model at a time, for every API key on the team. The setting can be on for every model, off for every model, or chosen per model. Switching a model off requires a confirmation that an exception under the applicable law applies, and that the customer still has any other duty to mark the output. New requests pick up the change within five minutes, and each opt-in or opt-out is written to the team audit log. Some models may be locked on, and those cannot be cleared in the console.⁵

Detection is statistical, and it is stronger when the text is longer. The detection API is offered only by application. The docs say the method slightly changes which tokens are chosen, and that it adds no characters and no hidden markup. The watermark applies to API text. It is not a new Grok model, and it is not a mark on consumer Grok chat.⁵