National Security – Competition – AI Geopolitics

Just as I'd started trying out Fable, an out-of-nowhere ban made me wonder: has AI stopped being merely an innovation arena where tech companies compete? Did it make you wonder too?

By the reviews, Fable was a very powerful model: it could interpret security vulnerabilities and hand the wrong(!) societies serious strategic capabilities. On that basis, a ban came down — like the export-license-controlled products of the defense industry.

According to the commentary, the core reason for the concern is SK Telecom — South Korea's largest telecom company and, at the same time, an investor and business partner of Anthropic — and its China ties, even though the company denies the claims. It's even said that Amazon, one of Anthropic's biggest investors and cloud partners, voiced concerns about possible vulnerabilities in Anthropic's newer models and whether their safeguards could be bypassed — and that this helped trigger the decision.

Which brings the question:

Should powerful AI models be treated as strategic technology, like chips?

If the answer is “yes,” we're looking at the first example of the restrictions people living outside the US will face from now on. And it won't be the last.

Another question that comes to mind: if an AI company works closely with the state while one of its powerful rivals has its model access restricted, how can we be sure the process is proceeding entirely independently of competition?

Anthropic has become one of OpenAI's strongest rivals today. Claude has found real success, especially with enterprise use, software development and a safety-focused approach.

On the other side, we know OpenAI has built more visible and strategic collaborations with the US government. This is, of course, a bit of a conspiracy-theory thought; but this isn't the first time Anthropic has been at odds with the US government, and it's telling that the period when it felt state pressure most coincides with the aftermath of OpenAI's cooperation deals with the government.

If the ban was brought solely for the sake of US national security, then this part is very clear:

In AI, power will now be measured not just by who develops the model, but by who can access it.

It seems that after the chip wars, one more term is being added to the dictionaries: “AI geopolitics.”