Maybe you all know about this:
The US government, citing national security authorities, has issued an export control order to stop all access to Fable 5 and Mythos 5 by any foreign person, whether inside or outside the United States, including Anthropic employees from other countries. The net effect of this order is that we should suddenly disable Fable 5 and Mythos 5 everything our customers to ensure compliance.
According to unconfirmed but possibly true reports, it was indicated that the model may be arrested. The released Mythos is already limited to bio questions and “AI development”, rather than actually, so now we return to the unavailable model.
Here are a few restrictions on the US government, not the only ones I might add:
1. Big companies need to stay in business. In addition, it wants their IPOs to go well. And now it is very difficult for the top companies to hire foreigners, which is a significant share of the top employees (Demis, Ilya, Andrej at the beginning!). It is also very difficult for large companies to conduct foreign business in a reliable and sustainable manner.
1b. How should American companies operating abroad use executive programs, moving forward?
2. It seeks to use model access as a tool for both hard and soft power, so model access must occur at some level. But it is very difficult to control what external agents will do with access to their partial model, when they get it in the future.
3. The US needs to stay ahead of China in the AI race.
4. The US needs to issue restrictions that are actually enforceable, and “US citizens only” doesn’t fit that bill. In addition (markets in everything!) it is easy enough to hire a traitorous American to access the tools to do wrong, or for that matter it is not difficult to deceive the nationality in various ways.
5. The USG cannot nationalize these companies and continue to run them successfully.
6. Chinese and other open source models are developing at some speed, even if they are now far behind the best proprietary models.
Is it a more likely scenario that the government tightens some of its programs and takes more precautions, and then allows the Mythos to be released again? Maybe with some extra protection?
Is there such a thing as a model that can’t be broken at all? I doubt that.
So we will be replaying this scenario over time, but all the time the companies and the government are in a very weak and critical position.
I’m willing to reject the philosophy of “safety” and bite the bullet in relation to it. As it is, these actions will not succeed in keeping us safe, including the reasons mentioned above.
In any case, these developments do not surprise me (they surprise me with their suddenness), since this type of approach is what governments have had for a long time now, the USG included or perhaps the USG in particular.
Escalation of the situation: Leopold, Aesop, and Mistral. National AI. Proponents of slow travel as a possible scenario. CEOs are quiet, quiet. As for China, will they jump at the chance, or at least be as scared as we are?
