The United States has decided to lift restrictions on the use of models based on Mythos, the most powerful artificial intelligence ever developed by Anthropic. It was released on June 9th but after three days Anthropic decided to shut down everything: according to the US government there were risks that its limits were being circumvented. Now it is available again, also in Italy.
The announcement came during the night, directly from him Anthropic account. Few words, exactly as happened with the past communications: “We have received notification that the Department of Commerce has lifted export controls for Claude Fable 5 And Mythos 5. We’ll start restoring access tomorrow and will share an update soon. We are grateful to our users for their patience and to all those who collaborated with us to redistribute the models.” The times are quite clear: tomorrow the users who had been able to use Mythos just for a few days they will be able to take charge of theirs again algorithms.
Fable 5 was the first commercial model of Anthropic based on Mythosa new family of algorithms that the market is looking carefully. In April the company led by Dario Amodei had presented Mythos, an artificial intelligence very different from those that are on the market. Something more powerful, so much so that it needs limits and security controls before being widespread. In its first phase it had only been released to a very small number of companies who were part of the Project Glasswinga project where Mythos was used for analyze and resolve bugs in the most critical software on the market.
The blocking of Fable and the risk of jailbreak, at least according to the United States
On June 9, Fable 5 became available to the market. It is the first model based on Mythos, although not entirely. To prevent its algorithms from being used for too sensitive topicsFable 5 was limited in responses. The model slows down and shifts: Instead of using Mythos, Fable 5 formulates its answers with alternative models such as Opus 4.8, also from Anthropic. Among the topics that had been limited were the cybersecurity and the medical biology. Three days after the stop: the United States blocks the spread of Fable 5. Or at least, they block the spread of this model for all users who are not citizens of the United States. A restriction that Anthropic can’t address by default, of course. And which therefore ends up blocking the model for everyone.
The case of Emma, the AI born in Italy that thinks of Hitler as an excellent president: “It’s experimental”
According to the ccommunications received at Anthropic the biggest risk was that of jailbreak: a procedure which, to put it very simply, serves to circumvent the rules defined by a software to push it beyond its limits. A risk that Anthropic has always denied. Now restrictions lifted, Fable 5 is back available to users. As we can read in the post published on the Anthropic blog, Fable 5 will be open all over the world. The company explained that in any case for this new release the security measures dedicated to model control.
Everything that was created in three days with Fable 5
In the few days that Fable 5 was available to the public they were published dozens of projects on social media. Complex sites created just by asking Fable for the code, video games replicated with a few prompts, model simulations on fluid mechanics, fully explorable 3D maps. Of course, given that the model was closed a few days later it is difficult to understand whether everything was done with just a couple of prompts. In any case, we will now be able to see the first impact of Mythos on the artificial intelligence market.
