OpenAI Caught a Fake Think Tank Faking Quotes From Noam Chomsky to Push Pro-Russia Propaganda

A think tank that never really existed just got taken apart in public, and the company doing the dismantling was the one whose product built it. Nvidia handed a startup 6 billion dollars for a piece of software rather than buy the company outright. A news and legal data giant built a competitive AI model for less than the price of a nice house. And in India, one team of engineers unveiled three jet engines built from scratch while another pair of athletes finally cracked a medal count that had sat frozen for fifteen years.

OpenAI took apart a fake think tank that quoted Chomsky and Fukuyama

In a report published this week, OpenAI said it banned a cluster of ChatGPT accounts tied to a Russian operation built around a fabricated think tank called the International Burke Institute. The site, registered in February 2025 and listing an address in Israel, presented itself as a serious policy shop and claimed prominent names such as Noam Chomsky and Francis Fukuyama among its roster of experts.

Neither scholar had anything to do with it. OpenAI reviewed 36 articles the site attributed to its supposed experts, published between September 2025 and May 2026, and found that 34 of them were lifted from elsewhere on the internet, sometimes with the wrong author’s name attached.

The operation’s centerpiece was something it called a sovereignty index, a made up metric that ranked countries in a way designed to flatter Russia and criticize Western governments. To get around the fact that ChatGPT is not supposed to be reachable from Russia, the people running the accounts connected through VPNs, wrote their prompts in Russian, and had the model draft posts for Substack, Telegram, X, Facebook and LinkedIn while stripping out anything that might give away where the writing came from.

Nvidia paid 6 billion dollars for software, not a company

Nvidia and the AI startup Poolside announced a deal this week that both sides went out of their way to describe as neither an acquisition nor an acqui-hire. Nvidia is paying 6 billion dollars to license Model Factory, the internal system Poolside built to train its own AI models, and separately investing 1 billion dollars in the startup itself at a valuation of 12 billion dollars before that money lands. That puts the full package at roughly 7 billion dollars.

More than 100 of Poolside’s engineers, 109 by Nvidia’s count, are joining Nvidia to work on Nemotron, its family of open weight models. Poolside’s three co-founders, including chief executive Eiso Kant and former GitHub chief technology officer Jason Warner, are staying behind to keep running Poolside as an independent company on separate research.

In a letter to shareholders, Kant and Warner framed the unusual structure as a matter of principle: they want a future built on artificial general intelligence to stay open rather than controlled behind closed doors by a handful of companies. Whether that holds up once the models in question are worth hundreds of billions of dollars is a different question, but for now Nvidia gets the tooling and the talent, and Poolside keeps its name on the door.

A 145 year old company built its own AI model for under half a million dollars

Thomson Reuters, the company behind Westlaw and Practical Law, launched its own proprietary language model this week under the name Thomson, the first in what it says will be a family of models carrying that name. Rather than train something from scratch, the company started with an existing open model, Alibaba’s Qwen 3.5, and retrained it on decades of its own legal and news content.

The company says it spent about 40 million dollars over two years on the people and computing behind the project. The final training run itself, thanks to efficiencies the team found along the way, cost roughly 450,000 dollars, a small fraction of what the frontier labs spend building a model from the ground up.

Thomson’s first job is inside CoCounsel Legal, Thomson Reuters’ AI assistant for lawyers, where it will handle Tabular Analysis, a feature built for reviewing large volumes of documents at once. It is a narrow, unglamorous debut for a model with a big ambition behind it: proof that a company sitting on the right proprietary data doesn’t need OpenAI or Anthropic’s budget to build something genuinely useful.

Indian engineers unveiled three home built jet engines in one afternoon

CSIR’s National Aerospace Laboratories unveiled three indigenous gas turbine engines on Tuesday at its headquarters in New Delhi, built to power drones, drone interceptors and small missile systems that India currently has to import propulsion for. The three engines, named NJ-05, NJ-50 and NJ-100, produce 5, 50 and 100 kilograms of thrust respectively, covering a range of small unmanned aircraft that until now had few domestic engine options.

Air Marshal Tejinder Singh, Chief of Integrated Defence Staff to the Chairman of the Chiefs of Staff Committee, attended as chief guest, and the CSIR-NAL team, led by scientist R Prathapanayaka, walked the room through the high speed turbomachinery and high temperature combustion work that went into building engines this small without sacrificing reliability.

It is not a headline grabbing announcement in the way a satellite launch is, but it closes a real gap. Miniature gas turbines are exactly the kind of component that gets bought off the shelf from a handful of foreign suppliers, and every one of those purchases carries the same quiet risk: someone else decides when you get the part and on what terms.

A badminton pair ended India’s 15 year wait for a world medal

At the BWF World Championships in New Delhi, India’s Treesa Jolly and Gayatri Gopichand won bronze in women’s doubles after losing their semi-final to China’s defending champions Liu Shengshu and Tan Ning 18-21, 21-13, 21-8 on August 22. It made them only the second Indian women’s doubles pair ever to medal at the World Championships, after Jwala Gutta and Ashwini Ponnappa took bronze at the 2011 edition in London.

The tournament, the sport’s biggest event outside the Olympics, ran at the Indira Gandhi Arena from August 17 to 23, the first time India has hosted it since 2009. Playing in front of a home crowd clearly mattered. Both players said afterward that reaching a medal match in their own country, in a discipline that gets a fraction of the attention singles badminton does in India, was exactly the kind of result they hoped would push more young players toward doubles.

Fifteen years is a long gap for a country with as deep a badminton bench as India’s has become. It says something that the wait ended not with one of the sport’s bigger names but with an unseeded pair nobody outside the circuit was tracking closely before the tournament started.

None of these five stories share much on the surface, but they all turn on the same question: who gets to decide what a piece of technology is actually for. OpenAI built a chatbot and someone used it to fake a research institute. Nvidia bought the machinery Poolside built without buying Poolside’s soul. Thomson Reuters proved a company doesn’t need to out spend the frontier labs to build something narrow and useful. And in New Delhi, on two very different afternoons, Indian engineers and Indian athletes both answered the same quiet question the same way: build the thing yourselves, and stop waiting for someone else’s part to arrive.