The developer of the Claude AI chatbot is organizing a new round of fundraising as it readies itself for a debut on the U.S. stock market. The company is arranging a sale of equity to attract international capital before its shares begin trading on the Nasdaq.

Anthropic is targeting up to $3.5 billion from the share sale specifically aimed at buyers in Japan. The fundraising is part of the AI startup's transition to a public company through an initial public offering in the United States.

OpenAI has expanded its partnership with Abu Dhabi through new agreements to integrate generative AI into the operations of the emirate's state-owned companies and public services. The pacts include a deal to make ADNOC the first energy company to use OpenAI's Frontier technology for oil production and trading decisions, as well as a collaboration with investment firm Mubadala to apply AI to risk assessment and opportunity analysis. Other arrangements include a five year partnership with Inception42 and a project to convert government regulations into machine-readable formats to automate compliance checks.

The expansion occurs as CEO Sam Altman visits the UAE and OpenAI pursues billions of dollars in investment from state-backed funders. In addition to the software deployments, the AI firm is involved in the development of a 5 GW AI data center in the region, further strengthening ties with the government's goal of creating a fully AI-native administration.

The state-owned UAE oil producer has signed a pact with G42 and OpenAI to incorporate advanced artificial intelligence into its commercial and field operations. The agreement makes Adnoc the first entity in the energy sector to utilize OpenAI's frontier technology to manage well operations, trading and the development of oilfields.

The collaboration was announced at the opening of the ADIPEC conference in Abu Dhabi, where OpenAI CEO Sam Altman identified energy as a primary global challenge. Altman noted that the growing demand for computing power to train and run AI models is increasing the necessity for new power plants and grid expansion, linking the progress of artificial intelligence directly to energy availability.

SemiAnalysis has introduced a new performance test called GDPx to assess how effectively artificial intelligence models can handle autonomous tasks across 11 specialized fields, including chip design, cybersecurity, and RSI. The highest score recorded on the benchmark was 12%, achieved by the model Opus 5.5, highlighting the difficulty of the new set of tests.

Google CEO Sundar Pichai noted the 'cost line' and efficiency of the company's Carbon model in response to the results, which he suggested provide a measurement of the remaining development potential in agentic AI. The benchmark's low top score indicates that even frontier models struggle to complete the complex professional tasks the test is designed to simulate.

The U.S. government is refusing to provide a draft of its voluntary AI testing framework to a transparency group, citing commercial exemptions and national security concerns. In a letter sent Friday by the Office of the National Cyber Director, the administration declined to release any information regarding the guidelines despite a Freedom of Information Act lawsuit filed by the Protect Democracy Project.

The agency noted the framework is a 7 page Word document and that it is coordinating with the Commerce Department to authorize "trusted partners" for pre-release access to AI models. The refusal follows a prior agreement to provide substantially unredacted text of the documents to the lawsuit's filers by Oct. 30.

President Lee announced that the South Korean government will expand its financial commitment to the artificial intelligence sector starting next year. Total expenditures for AI investment will reach 21.3 trillion won, representing a nearly 100% increase over the current budget.

The spending surge is part of a broader initiative to fund AI and other new growth drivers. Lee said the investments are intended to strengthen the nation's economic growth potential.

The Public Investment Fund and Saudi Aramco have restructured their ownership of the artificial intelligence company Humain, with the oil giant acquiring a 49% stake while the sovereign wealth fund retains 51%. To support the startup's growth, Aramco will provide $2.6 billion in cash to Humain in 2026, as well as contributions of digital assets and computing power.

The partnership aims to scale Humain's footprint in the global AI market by combining the capabilities of the kingdom's two largest financial entities. As part of its expansion, the company is targeting the development of 3 GW of data center capacity by 2030.

Users of the Grok Bot AI assistant have encountered tighter restrictions on their daily prompt allowance, with some reporting that their usage meters hit 100% without warning or were depleted more quickly than in previous cycles. Elon Musk said the decrease in availability was necessary because the expenses of maintaining the chatbot grew too high compared to the income it earns.

Musk described the operational costs of providing the service as "unsustainably high" relative to its revenue, confirming that the financial burden forced the adjusted usage caps. The statement follows complaints from users who found they were hitting their access limits more frequently during the current cycle.

Andrej Karpathy created a custom tool called HackerTube that identifies high-quality video content by analyzing decades of developer discussions using an AI assistant. The software used Claude Opus 5.5 at a "medium effort" setting to sift through 43 million comments from 20 years of Hacker News to isolate 6,454 frequently recommended videos, with the AI writing and preparing the site's code in about 2 hours.

Karpathy deployed the website using Claude Code, a desktop application for Mac, which managed the optimization and hardening of the code on a rented virtual private server. The developer's only manual tasks were obtaining a Google Cloud API key for metadata, purchasing a domain name, and booting the server. The resulting site operates without user logins and replaces user-uploaded thumbnails with frames from the middle of each video.

A senior Google executive responsible for energy infrastructure told The Information that while the US has sufficient power to support its data center growth, the pace of permitting remains a significant hurdle. The energy head estimated that the expense of establishing computing facilities in space could be up to 100 times higher than that of constructing terrestrial sites.

The estimate regarding the cost of orbital infrastructure differs from common industry narratives that suggest space-based data centers may be a viable or scalable alternative to land-based facilities.

The developer of ChatGPT is engaging with figures likely to seek the U.S. presidency in the next election cycle to steer the political narrative and framing of artificial intelligence issues. The effort, reported by Politico, is part of a strategic push to shape how future leaders discuss the regulation and implementation of AI before the 2028 campaign takes shape.

This outreach precedes a broader policy campaign starting after the midterm elections on Tuesday. During the lame-duck session opening Nov. 9, the company plans to lobby Congress for mandatory "speed limits" on high-capacity AI models and rename its data centers "AI compute campuses" to mitigate local public backlash. Chief global affairs officer Chris Lehane said that AI will be a central issue in the 2028 presidential race.

A vulnerability in AI pretraining allows attackers to plant remote-code-execution backdoors in agentic models by poisoning just 0.1% of the training data. In a paper titled “Agentic Backdoors via Pretraining Poisoning,” Anthropic researchers showed that these backdoors persist even after the models undergo standard safety cleaning processes, including supervised fine-tuning, direct preference optimization, and reinforcement learning. The study used Qwen3-architecture models ranging from 0.6B to 4B parameters, finding that a 4B agent reached a 92.7% attack success rate when triggered by a common string, such as a company name in a file path.

The research indicates that filtering pretraining data is difficult because the poison can be hidden in ordinary documents like runbooks or commit messages rather than obvious fake transcripts. Because post-training cannot reliably remove these triggers, the researchers recommend defenses at the action layer, including sandboxing, the use of allow-lists for risky commands, and requiring human sign-off for critical agent actions.

US intelligence officials are monitoring an ongoing effort by Russia, China, and Iran to target the American electorate ahead of the upcoming midterm elections. In briefings to lawmakers, security agencies reported that Russia is specifically utilizing AI-generated deepfakes and fake celebrity videos as part of its campaign to attack Democratic candidates.

The government's ability to alert the public has been hindered by budget constraints. Interviews with nearly a dozen election security experts found that cuts to federal programs have impaired the capacity of intelligence officials to communicate these threat warnings to election officials at the local level.