Anthropic: Kremlin propaganda used Claude to “transform” news from Romania and Moldova

The Threat Intelligence report published by Anthropic describes an operation in which Claude was integrated into media production workflows linked to Russian state media. According to the report, a former Sputnik Moldova editor-in-chief used Claude to transform Romanian and Moldovan news, polls, and opposition posts into Russian-language materials, later published on channels such as Sputnik Moldova and RIA Novosti. Anthropic says some of these materials were amplified across multiple publications and channels to create the impression of independent confirmation. The report also mentions the use of fabricated and defamatory claims about the President of Moldova, Maia Sandu, ahead of the 2025 parliamentary elections.

Why it matters: the case shows an essential shift in how propaganda can be used. Claude was not just a tool for drafting text, but a component of an editorial workflow capable of taking real information from Romania and Moldova, translating it, reformulating it, and adapting it to a predetermined political narrative. Anthropic says it identified and blocked the accounts involved and passed information on to partners and authorities.

Anthropic CEO Dario Amodei calls on the AI industry to slow down the development of frontier models

Anthropic CEO Dario Amodei has called on the AI industry to slow the pace at which frontier models are developed. In an essay published on September 12, he does not propose halting progress, but rather a more controlled pace, based on independent evaluators, cooperation among AI labs, and closer collaboration between companies and governments. The call comes immediately after Anthropic published an extensive report on the abusive use of Claude models in cyber operations, surveillance, fraud, political influence, and weapons development.

Why it matters: Amodei is not speaking from outside the industry, but from the leadership of one of the most important AI labs. The message is all the more relevant given that Anthropic has just documented cases in which advanced models were used in real-world operations.

Nvidia could invest up to $10 billion in Anthropic's IPO

Nvidia is negotiating to take part as an anchor investor in Anthropic's upcoming initial public offering, with the amount under consideration potentially reaching $10 billion. According to Reuters, Anthropic is weighing an offering that could raise up to $100 billion and value the company at roughly $2 trillion. Discussions are not finalized, and the terms may still change.

Why it matters: this is one of the more interesting economic relationships in the new AI industry: Nvidia is simultaneously a supplier of infrastructure for Anthropic and a potential investor in the company. Anthropic needs enormous amounts of computing power, and Nvidia benefits from this expansion both by selling chips and, potentially, through the rising value of its stake. It's an example of how the AI ecosystem is starting to resemble its own financial infrastructure, an idea AIdapted has already tracked in its analysis dedicated to the industry's major rivalries and massive funding race.

California adopts a tough package of rules restricting technologies used by children

California Governor Gavin Newsom has signed a package of 13 laws that introduces new restrictions on technologies used by children. Social media platforms will no longer be able to offer users under 16 certain features considered addictive, such as infinite scroll and autoplay. The new rules also introduce requirements for AI chatbots, parental controls, and safety assessments, as well as restrictions on toys that integrate AI chatbots.

Why it matters: California is beginning to treat social media and AI chatbots as two sides of the same problem: systems capable of influencing children's behavior and creating highly persuasive relationships with users. The legislation will be closely watched because California is one of the world's most important tech markets, and the rules adopted here can drive product changes for companies operating globally.

Revolut confirms a breach caused by fake government requests

Revolut has confirmed that it disclosed sensitive customer information after attackers submitted requests that appeared to come from a legitimate government institution. The company notified affected customers and the relevant authorities. The incident shows that the attack did not rely solely on breaching an IT system, but on manipulating the internal procedures the company uses to verify and respond to official requests.

Why it matters: this is a very clear example of social engineering (manipulating people into taking an action that technical systems would not allow directly). For companies, the problem is becoming more complicated as AI can quickly produce convincing official messages, documents, and digital identities. In an attack like this, the most vulnerable component can be the very employee trying to follow security procedures.

OpenAI launches ChatGPT for the financial sector

OpenAI has launched ChatGPT for Financial Services, a version built for financial institutions in collaboration with Morgan Stanley and Evercore. The product combines OpenAI's models with financial data supplied by companies such as Daloopa, PitchBook, and LSEG News, and offers features for research, financial modeling, earnings analysis, and client document preparation. An important feature is the ability to trace claims and figures back to their sources.

Why it matters: for the financial industry, a chatbot that “knows a lot” isn't enough. Analysts need to be able to verify where a figure comes from and demonstrate how a conclusion was built. For Europe, where financial services are heavily regulated, this direction is especially important.

OpenAI says its infrastructure serves over 1 billion people a week

OpenAI has published details about Habitat, its internal online storage platform that powers ChatGPT, the API, and other services. The company says the infrastructure serves products used by over a billion people a week and must handle tens of millions of requests per second worldwide. Habitat evolved from a simple software library into a globally distributed platform as OpenAI's user base grew.

Why it matters: the figure of over a billion changes the perspective on AI. Large systems need to be able to run continuously for hundreds of millions, or even billions, of people.

AI is being used to search for new antimicrobial molecules

Researcher César de la Fuente and his lab are using ChatGPT and Codex to search the genomes of living and extinct organisms for molecules that could become new antimicrobials. The research aims to speed up a stage that traditionally can take years: identifying candidates worth testing later in the lab against treatment-resistant microorganisms.

Why it matters: this is one of the examples where AI can deliver very concrete value beyond software and productivity. Models can rapidly analyze vast spaces of possibilities and help researchers narrow down the list of molecules worth investigating experimentally. AI doesn't automatically turn a molecule into a drug, and results still need to be validated biologically, but it can accelerate one of the most difficult stages of the discovery process.

Z.AI seeks $5 billion to expand AI infrastructure in China

Chinese company Z.AI has launched a financing operation worth approximately $5 billion, through a roughly $2 billion Hong Kong share placement and about $3 billion in convertible bonds. The money is intended to expand the company at a time when Chinese labs are trying to rapidly scale up computing capacity and model development.

Why it matters: the AI race is increasingly becoming a race for capital, chips, and infrastructure. Z.AI isn't just trying to build better models, it's trying to raise billions for the physical capacity needed to train and run them. Alongside the huge investments in the US, deals like this show that China is also building its own financial and industrial ecosystem for AI, which makes the competition between the two blocs harder to reduce to a simple comparison between models.

Cohere in talks for up to $3 billion in funding, with Canadian and German participation

Canadian company Cohere is in advanced talks for a funding round of between $2 billion and $3 billion, at a valuation of roughly $20 billion, according to information published by The Globe and Mail. The Canadian government and existing investors are expected to take part, while the German government is also discussing a possible investment. The round could become the largest funding raised by a Canadian startup.

Why it matters: the Cohere case is interesting for Europe because the Canadian company has already merged with Germany's Aleph Alpha, building a bridge between North American capital and research and the European ecosystem. If Germany does end up investing directly in the new round, governments would be moving beyond funding AI infrastructure to also capitalizing the companies that develop the technology. It's another form of “AI sovereignty”: not necessarily developing a model exclusively in Europe, but retaining European influence over the companies and strategic infrastructure.

The Data Act enters a new phase for connected products

Starting September 12, an important phase begins in the application of the Data Act provisions concerning data generated by connected products and related services. For certain products placed on the market after this date, they must be designed so that the data they generate can be accessed by the user in a format that is easy to use, secure, structured, and, in principle, free of additional costs for the access provided for under the regulation.

Why it matters: for AI, data generated by machines, vehicles, industrial equipment, smart devices, and other connected products is a strategic resource. The Data Act aims to change the relationship between manufacturer and user: data produced by a device is no longer automatically treated as a resource controlled exclusively by the company that made it. In the long run, this could make it easier for AI services to emerge that analyze data from millions of connected products, and could reduce dependence on the platforms that control this data today.

Sources:

●        Dario Amodei — essay “We must pace the frontier”

●        Anthropic — Threat Intelligence Report, September 2026

●        Reuters — Nvidia in talks for Anthropic's IPO

●        Reuters — California adopts new rules for social media and children

●        Reuters — Revolut confirms data breach

●        OpenAI — ChatGPT for Financial Services

●        OpenAI — Habitat: storage infrastructure

●        OpenAI — Using Codex/ChatGPT to search for antimicrobials

●        Reuters — Z.AI launches a $5 billion financing operation

●        Investing.com / The Globe and Mail — Cohere raises up to $3 billion

●        European Commission — Data Act

●        AIdapted — Major AI rivalries: former colleagues, billions of dollars, and an unprecedented race