presses

The United States has turned off the tap most powerful, expensive and advanced AI model ever released by Anthropic, but it has done so in a disconcerting move and with small print. Yesterday, June 12, the White House ordered Anthropic to immediately shut down Fable 5 and Mythos 5 for anyone who is not a US citizen. The order arrived without giving too many explanations about its duration or technical justification, according to Anthropic’s public statement. This government intervention marks a before and after in the modern technology sector: they are not restricting the sale of chips (a common practice), but vetoing access to general access software as a matter of national cybersecurity. What’s happening with Fable 5 and Mythos 5. The US government has issued an export control directive suspending all access to Fable 5 and Mythos 5 for anyone who does not have US citizenship, whether inside or outside the United States. This includes non-U.S. Anthropic personnel. In order to adhere to the directive that came into effect immediately, the company led by Dario Amodei has had to deactivate both models for everyone. At the time of publishing this article, the notice appears when trying to access it. According to Anthropic, the government believes it has discovered a way to jailbreak the model. The company does not agree: after reviewing the demo of this technique to deceive the model, it concludes that the vulnerabilities found are minor, were already located and applicable to other models on the market and can be identified without the need for any circumvention technique. What you see right now if you enter Claude Why is it important. Because Mythos It is a specifically designed model for cybersecurity and its blackout affects both commercial users and those entities in charge of active defense management that are using it. With this measure, Washington turns its cutting-edge AI software into a national security asset. It is worth remembering that Anthropic is already on the blacklist of the Pentagon, so he considers it too dangerous for his Government to use. It is also now too dangerous for foreigners to use. Besides, as Anthropic points outsets a dangerous precedent: if the criterion for withdrawing a model is that someone finds a vulnerability, in practice none will be able to be launched because the company that releases software without a single failure raises its hand. That is to say, if the White House makes this exceptional movement the norm, it will be a shot in the foot to develop cutting-edge models: slower launches due to this “impossible” debugging process, personnel decimated by not being able to count on foreign specialists and if they cannot be marketed abroad there will be less income. Context. This order is the latest chapter in a soap opera of disagreements between Washington and Anthropic that dates back to the beginning of the year. In March, the Pentagon considered that the company was a “supply chain risk.” In its statement, Anthropic has verified that that level of capacity that the government identified as dangerous in Fable 5 is already available in other models on the market, including the GPT-5.5 by OpenAIearlier and more widespread than Mythos. However, that version of ChatGPT has not suffered any suspension. The asymmetry of treatment between OpenAI and Anthropic is evident. The geopolitical context is also important: the United States and China are immersed in a technological race unprecedented in recent decades and each power is playing its weapons, from tariffs to critical materials like rare earths to vetoes on chip sales, EDA software or the Export Control Reform Act 2018, where new technological categories fit, such as the most advanced AI models. Of course, for the Asian giant, every obstacle has worked as a kind of catalyst to advance faster and be increasingly independent of outside technologies. In detail. The order’s enforcement mechanism is supported by the “export control directive” managed by the Department of Commerce. In practice, it means that accessing Fable 5 or Mythos 5 as a foreign citizen (even if you are in the United States) is grounds for infringement. Anthropic tried to anticipate the problem with thousands of hours of testing with the US government, shared it with 40 organizations that manage critical infrastructure and then, with another 150 entities more precisely so that they could find and thus be able to correct vulnerabilities before a third party with malicious interests did so later. It hasn’t been enough. Because Mythos It is a specifically designed model for cybersecurity. Before its public deployment, Anthropic shared it with organizations that manage critical infrastructure precisely so that they could find and thus be able to correct vulnerabilities before a third party with malicious interests did so. Its blackout affects both commercial users and those entities in charge of active defense management. Yes, but. For now, Anthropic complies with the order, but makes it clear that it does not agree and that it is working to resolve it: “We believe this is a misunderstanding and we are working to restore access as soon as possible.” The company behind Claude says it supports the government in blocking really dangerous technologies, but that this process has to be transparent, fair and based on real technical facts. In short, he has not said the last word. On the other hand, the million-dollar question is whether the US government can apply the export control law, initially designed for chips, satellites or critical and specific software, on an AI model available on the internet and in general use. In fact, this regulatory movement puts on the table the importance of having a clear law that defines when, how and with what guarantees a government (in this case, the United States) can intervene. In Xataka | Spain does not want to wait for Mythos to wreak havoc on its companies: that is why it is going to ask for “early access” to Anthropic In Xataka | Claude Fable 5 is the most powerful public … Read more