/ Government Blocks Fable 5 access after Export control order
Letty Simone is an expert AI writer. She Covers AI news, reviews tools and updates the audience with the latest AI updates. She joined TheTweaks as an AI writer but Prior to TheTweaks she worked as an AI product tester at a business software company. She thinks that the majority of AI reporters represent the story wrongly and she has an aim to do it in a better way.
AI research firm Anthropic has been compelled to disable access to two of its advanced AI models, Claude Fable 5 and Mythos 5, following a last minute government order requiring the firm to prohibit any usage by foreign nationals as of today. Specifically, the instruction was issued on Friday at 5:21 PM ET to limit access of these models to anyone foreign either inside or outside the US territory.
Anthropic doesn’t have the ability to check users’ nationality instantly, the company decided to disable access globally for everyone, causing heated discussions around the first ever large scale government action targeting the accessibility of frontier tier AI models.
The directive has been sent out by the US Secretary of Commerce Howard Lutnick via his letter to Anthropic CEO Dario Amodei, putting both Fable 5 and Mythos 5 under export control.
The conditions set out by the directive include:
Since the Anthropic system is unable to verify the nationality in real-time, the decision was made to totally shut down both models in the company’s infrastructure.
Consequently, the following measures were taken:
Thus, the action resulted in a total shutdown of frontier tier models of the company.
According to insiders, the government’s decision has been made following another company claiming that the Mythos 5 can be jailbroken with the ability to perform certain operations.
Those included the tasking to make the model:
While this type of capability is inherent to many AI models used in debugging and cybersecurity related research, the administration viewed it as a serious national security issue for more advanced reasoning engines like Mythos 5.
Officials were not worried about jailbreaking as such, but rather the possibility that those systems can be used to:
Additionally, it is believed that the US government tried to block the launch of Fable 5 until it was released following the company’s safety evaluation.
The company has disagreed with the government’s assessment, saying that this measure is excessive.
“We do not believe that a potential jailbreak of one of our safety classifiers is sufficient reason to recall an otherwise commercially available model which would go live shortly and will be used safely by hundreds of millions of people. Should the same standard be applied throughout the industry, it would effectively stop the deployment of all new models by frontier AI companies.”
The models went through thousands of hours of testing;
The above models belong to Anthropic’s latest series of reasoning engine AI models. Although both belong to the same architecture, they are distinguished by the level of safety precautions.
Another very important technical aspect behind that decision is enforcement. As the AI models are deployed globally and users connect from anywhere, it is technically impossible for Anthropic to monitor usage by nationality in real time. Therefore, there has been no other way to implement the government directive besides shutting down the models completely. What follows from here is that the regulatory requirements for AI companies are starting to develop faster than systems themselves.
Timing wise, this decision comes at the worst possible moment for Anthropic. The company has recently filed its confidential IPO prospectus with the SEC estimating the total value at $965 billion while the annual revenue run rate stands at $47 billion. This document was delivered earlier this month. On top of that, regulatory pressure on the firm is growing significantly. Not long ago, the US Department of Defense marked Anthropic as a threat to the national supply chain security, normally applicable to entities which are sensitive from a national security standpoint. Thus, a combination of upcoming IPO, government concerns, and frontier model ban create one of the most turbulent regulatory periods in the history of AI industry.
The reaction from the industry is rather mixed. Proponents of that decision state that frontier AI models should definitely be under strong control by national security authorities in order to ensure safety for the whole population. Critics say that the definition of “potential risk” is overly inclusive and might impede the pace of innovation globally. Also, the same measures might be used in the future regarding any other AI company’s model releases.
The main problem, however, does not lie in shutting down the systems but in setting the precedent for how future frontier tier systems will be controlled around the globe.
From TheTweaks perspective, what we see happening here is a turning point in the whole AI industry. Never before the government has been taking this type of step involving model access restrictions, leading to global shutdown. Most importantly, we see a new direction in the regulation which bases upon potential capabilities of a system rather than any misuse. The future of AI systems might include non uniformity between regions, license-based usage restrictions, and even classification according to the governmental risk assessment framework.
At TheTweaks we view this as the beginning of a new era of AI where controlling the AI becomes as important as building AI itself.
It is safe to assume that depending on the result of Anthropic’s legal battle, the model release and licensing process in the future will undergo significant changes.
Anthropic promises to give further technical details within 24 hours. We’ll provide updates to this breaking news story as soon as it is relevant.
Tim Cook’s 15 years as Apple’s chief executive officially ended on September 1 and now the torch has been passed on to John Ternus, who is the company’s long-time hardware. It’s Apple’s first leadership change…
.jpg&oldid=1142454924){kind=link}
Be respectful and constructive. Have a question or feedback? We’d love to hear from you. Contact us at contact@thetweaks.com