The Anthropic Fable saga is so entangled in the messy centre of tech and politics that it is a challenge to see a clear narrative on which to hang the story. But it is important because it is a version of what was once called the military-industrial complex. (Dwight Eisenhower coined the phrase in 1961, and it was meant as a warning.) AI has redefined the meaning of the phrase, and the Fable affair is its poster-boy.
For those who missed it, here is what happened. It is early April and Anthropic is about to release its latest and much-anticipated AI model Mythos to the public. But it quickly becomes apparent that it’s so powerful that it could also be used to hack things that were considered un-hackable, including systems safeguarding US national security.
Everyone freaks out.
Anthropic delays the public release of Mythos and shares “Mythos Preview” with a select few big tech companies like Microsoft, Google, Oracle and others, giving them time to hack-and-patch their systems before going to the general public. The US government is kept informed every step of the way, notwithstanding that Anthropic and the Trump admin do not like each other very much, having incompatible ideas about regulation.
That should have been the end of it. But a few weeks later, on June 9, Anthropic releases a somewhat sanitised version of Mythos called Fable, supposedly stripped of the most dangerous edges (like hacking prowess) but including all of Mythos’s other shiny buttons (which are blinding enough to have been launched into first place in AI benchmark races). But within days, techies from Amazon (which is, ironically, an Anthropic investor) report that they have found a ‘jailbreak’: basically, a way to get around Fable’s newly erected barriers and guardrails.
They tell Amazon CEO Andy Jassy. Jassy calls the White House (among others).
Everyone freaks out again.
But this time, a panicked and under-informed White House jerks its knee with prejudice. They order Fable to be disabled for everyone but US nationals, via a hastily written export control directive. They also give Anthropic 90 minutes to do this. This is lamentably stupid. It means that in practice, some of the software engineers (there are many non-US tech staff in all AI companies both in the US and elsewhere who helped build Mythos/Fable) are suddenly proscribed from using the very software they designed. It also blocks US allies from using the system. It ignores the fact that other frontier models like ChatGPT can do much the same thing as Fable. And it ignores the fact that the ‘jailbreak’ was so narrow and constrained that it barely qualifies as a jailbreak.
But alas, those were not the main problems.
The consequence of the White House action is that the world saw that access to cutting-edge AI could disappear overnight because of a policy decision made by just a few people in Washington. For the United States government, this may have seemed like a reasonable (if hasty) national security precaution.
For everyone else, it was a wake-up call.
Imagine you are India. You have a population of 1.4 billion people, one of the world’s largest software industries and ambitions to become a global AI power. Yet your developers wake up one morning to discover that access to a critical technological platform can be switched off by a foreign government. Or imagine you are Saudi Arabia or the United Arab Emirates, spending tens of billions of dollars on AI infrastructure to diversify beyond oil. Or South Africa, Kenya or Nigeria, hoping to leapfrog stages of economic development through AI-powered education, healthcare and financial services.
The obvious conclusion is not that you should become more dependent on American AI. It is that you should build your own (expensive) system, or (more likely) buy open-weight systems only. Which means, with a few exceptions, that you will turn to China. Why this was not entirely obvious to the White House is a little hard to understand.
Every developer in India who chooses a Chinese model because an American one might disappear is a tiny geopolitical loss. Every African startup that builds its products around an open-weight ecosystem rather than a proprietary US API shifts the centre of gravity towards the East. Every Gulf sovereign wealth fund that decides to back domestic AI laboratories instead of relying on Silicon Valley is a hedge against American leverage.
There is an irony here that borders on the tragic.
The United States has spent years trying to limit China’s rise through semiconductor sanctions and export controls. The theory is straightforward: deny your strategic competitor access to the best technology and you preserve your lead. But software and mathematics are almost impossible to protect. This was famously demonstrated by the failed government attempts to legally ban encryption algorithms during the second half of the 20th century: a saga finally put out to pasture in 1999.
The US may have successfully hobbled China’s AI ability to import the most advanced Nvidia chips, but AI is not hardware. Unlike advanced chipmaking equipment, AI capabilities diffuse rapidly. Researchers move. Papers are published. Techniques are replicated. Open-weight models improve. And in spite of the fervid imaginations of only-in-America types, US engineers and developers are not more intellectually endowed than citizens of other countries.
More importantly, Chinese laboratories have increasingly embraced openness as a strategic advantage, releasing powerful models that can be downloaded, modified and deployed almost anywhere, resulting in the global diffusion of soft Chinese power. American analysts have warned that this may itself be a geopolitical strategy. If Chinese AI becomes broadly accessible while America goes off its own, Beijing could become the default partner for countries seeking technological autonomy.
If that happens, Washington’s actions against Anthropic will have achieved exactly the opposite of what was intended.
Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg and a partner at Bridge Capital and a columnist-at-large at Daily Maverick, Daily Friend and Currency News. His new book “It’s Mine: How the Crypto Industry is Redefining Ownership” is published by Maverick451 in SA and Legend Times Group in UK/EU, available now.
[Image:ChatGPT]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend