- Статьи
- Economy
- Models in a row: the ban on Anthropic AI will stimulate the development of such software worldwide
Models in a row: the ban on Anthropic AI will stimulate the development of such software worldwide
The closure of access to the latest Anthropic AI due to US requirements is described by high-tech market participants as an unprecedented event. In their opinion, the ban will give impetus to the development of such tools around the world. Representatives of the EU and Canada have already announced the need to step up such work. The situation around Anthropic has become an additional argument in favor of strengthening digital sovereignty and developing its own technologies, experts say. At the same time, government interference in the activities of a private company can raise questions among investors of Anthropic against the background of preparations for an IPO, they believe.
Why was access to Anthropic AI products blocked?
There have been no cases in the short history of the mass market of artificial intelligence when the state has directly restricted access to AI products, experts interviewed by Izvestia believe. Anthropic released the latest language models Claude Mythos 5 and Claude Fable 5 last week, and access to them was closed on June 12. The US government has demanded that the manufacturer ban access to them for all non-American citizens, including Anthropic employees. American media sources claim that the Commerce Department gave 90 minutes for this. The company found this requirement impossible and completely blocked access to these resources.
The Mythos 5 and Fable 5 AI models are configurations of the improved Claude 5 base model, with Mythos designed for a limited range of corporate clients, and Fable is its custom version. According to Anthropic, Claude Mythos 5 is able to identify software vulnerabilities better and faster than a human. Both configurations have an unprecedented level of protection compared to other AI models. Mythos 5 was developed in collaboration with the US government. The user version of Fable 5 includes a ban on instructions for performing dangerous actions, access to dual-use information, and "adult" content.
According to American market participants, restrictions could have followed after Amazon, one of the shareholders of the AI developer, analyzed the Claude Mythos 5 and Claude Fable 5 models and prepared a special report. The document was not published, but experts believe that the company considered the new systems to be a potential tool for hackers and other cybercriminals, which poses risks to national security. Claude Mythos 5 and Claude Fable 5 provide protection against dangerous requests related, for example, to software hacking, weapon design, poison synthesis, and the like. But, as Axios previously reported, immediately after the release of the new models, information appeared about the creation of tools to bypass these filters, which became the reason for restrictions by the US authorities.
— These models can, for example, find such gaps and vulnerabilities in operating systems that humans cannot find. Or, for example, to restore the source code of the program, which was considered lost and now such features are available to anyone," said a source of Izvestia in the IT market.
The situation with the closure of a commercial product that has already been released on the market at the request of the state is truly unprecedented, says Vladimir Zykov, editor-in-chief of the IT publication Runet. The situation with Anthropic may become a reason for real steps in the field of strengthening digital sovereignty in many countries of the world, he did not rule out.
According to the founder of the Internet Search company (part of the Cybersystem Group of Companies) Igor Bederov, we are talking about a tectonic shift in global digital politics.
— We are witnessing a transition from the era of unrestrained technological globalism to the era of rigid digital mercantilism. This is an indicator that artificial intelligence has finally moved from the category of commercial products to the category of strategic deterrence tools," he said. — We have seen attempts to slow down TikTok, Clearview AI trials due to total biometrics, but these were targeted attacks on data collection or supply chain security. Here, for the first time, the state is hitting the level of cognitive abilities of a commercial product that has already been released to the market. This can be called unprecedented from the point of view of the mechanism, since the US administration actually applies export control standards to the neural network program code.
According to the expert, the closure of access to Anthropic technologies is a powerful argument in favor of strengthening digital sovereignty. Russia should not find itself in a situation where critical infrastructure depends on a "black box" that can be turned off from overseas with a single call from the State Department, he is convinced.
Izvestia sent requests to the Ministry of Finance and Anthropic.
What will replace the American language models
A ban on the use of Anthropic models outside the United States could have far—reaching consequences around the world - first of all, it is important how China reacts to this, said Karen Ghazaryan, director of the Institute for Internet Research. On the one hand, if Chinese developers are not catching up yet, they are actively reducing the gap in the field of AI. On the other hand, there are signs that Beijing is seeking to restrict other countries' access to its AI developments. For example, the authorities recently imposed a ban on the sale of one of the startups specializing in artificial intelligence, the expert recalled.
The ban on the use of Anthropic products has already raised concerns in Canada.
"The situation that we are in together now because of Mythos and Fable could happen again if we rely too much on certain models,— said Prime Minister Mark Carney. "No one has done anything wrong. But we will do the wrong thing if we simply accept, do not learn a lesson, do not expand and diversify our capabilities.
European politicians also called for accelerating the development of AI. Countries that do not urgently start developing their own AI models will increasingly depend on the decisions of other states, said the leader of the French National Unification Party, Jordan Bardella.
Theoretically, the US government's ban on exporting the most advanced AI models should lead to a parade of national AI sovereignty, but in reality, only a few countries can achieve this, said Leonid Konik, a partner at ComNews Research.
"This requires too many resources: scientific personnel, IT specialists, powerful data centers with graphics processors and electricity to supply them," he calculated.
What happened to Anthropic is the first high—profile precedent when the state turned off access to advanced AI, using it as a tool of geopolitical pressure, says Leonid Drobyshevich, an NTI technology expert.
"A signal has already been sent to the market: linking to one supplier from one jurisdiction is a strategic risk, and corporate clients around the world will now be looking for multi—vendors and local alternatives," he said.
However, according to him, it is impossible to quickly achieve parity in the AI market. There are actually only two players on the frontier today — OpenAI and Google DeepMind. They are followed by xAI, China's DeepSeek and Qwen, and in Europe, Mistral is the closest to the leaders. At the same time, as the expert notes, none of the non-American developers have yet been able to close the gap with the leading closed models.
Nowadays, AI technologies are becoming a key tool in global competition. The few states that will be able to create sovereign AI models will de facto dictate the terms to the rest, experts say.
Meanwhile, in early June, it became known that Anthropic had filed a confidential application for a public offering of shares with the U.S. Securities and Exchange Commission (SEC).
According to Igor Bederov, the impact of the situation on the IPO of Anthropic is rather negative. The company builds an investment story around growth prospects, but regulators are actually signaling to investors about the potential limitation of plans due to geopolitical barriers. This, in his opinion, can significantly reduce the multipliers and capitalization estimates. If the company transforms into the most closed structure and becomes a de facto government-oriented project, it can grow in value as a strategic asset. However, this will lead to a decrease in liquidity and transparency for the public market, the expert notes.
On the one hand, government restrictions on a particular business always cause investors to be wary, Vladimir Zykov added. At the same time, the very situation with "closed" models, in his opinion, may indicate a high assessment of the company's technologies and its potential for technological breakthrough, which is a positive signal for the investment community.
Переведено сервисом «Яндекс Переводчик»
